Does ChatGPT Trust Itself?
Businesses And Brands Beware
As the public begins to form its relationship to Artificial Intelligence, there are more questions than answers. For every organization who utilizes AI, one of the most critical questions that all stakeholders will intuitively ask is “can we trust it?”. Can leaders trust the outputs from their own models to make critical decisions? Will customers trust themselves if the brand is predicated on AI? Will the efficiencies AI brings outweigh the unforeseen consequences?
It is easy to overlook that trust is the foundation of our relationships at every level of life - from our personal relationships, to our monetary systems, to the institutions that make up our society. Already, trust in our institutions, businesses, political parties, etc…are at an all time low. Where trust is missing, we witness the breakdown of systems and their ability to function. For brands, a loss of trust can seriously damage or kill the business. Deploying AI significantly raises these risks.
Artificial Intelligence has some unique features that have no historical precedent. As Historian, Yuval Noah Harari points out, “AI can make decisions by itself. The second thing everybody needs to know is that this is the first tool in human history that can create new ideas by itself.”
At this dawn of the AI era, it is worth looking at the topic of trust and our relationship with this new technology. Even more importantly (and for the first time in history) we need to inquire about AI’s relationship to itself.
ChatGPT's meteoric rise has been astonishing, with myriad experiments pushing its boundaries. Numerous enterprises, including us at Optimal Trust, are keen to leverage AI's might. We introduced ChatGPT to our Optimal Trust Model - a framework designed by Stuart Diamond, initially for healthcare providers aiming to amplify trust. The model later evolved as a trust quantifier and training program for top-tier financial institutions. We use the Optimal Trust Model to quantify trust in organizations and to teach people an intuitive and effective access to building trust in the workplace, with clients, and in their personal lives.
While we normally focus on organizational development, training leaders, advisors, and high level sales people, we were curious to see what would happen when we asked ChatGPT to use our model to assess itself through the primary components of trust. The results were fascinating. We have posted the prompts and responses below.
Background on the Optimal Trust model
The model consists of 2 axes (see below). The X axis contains the 6 most universal components of trust (Competency, Aligned Interests, Perceived Intentions, Quality of Communication, Shared Values, and Integrity). The Y axis is what we call the “Levels of Self,” or the way that these components may occur and impact us differently depending on the situation (i.e. how competent I am at my job, how competent my organization is, how competent the industry is….) This is the model we train individuals in and is the model we asked ChatGPT to use to score itself.
Below is the exchange we had with ChatGPT. You will see that we asked the AI to assess itself through the components of trust, first. Then, we have it assess OpenAI (Level of Organization) through the components or trust and finish with assessing the overall AI industry (Global) through the same components. (Interactions highlighted in Blue)
Prompt to ChatGPT:
We would like ChatGPT to evaluate itself using the Optimal Trust Model, which employs an innovative framework for measuring trust, both qualitatively and quantitatively. The framework uses a grid, which can be as simple or complex as needed and is easy to memorize.
The grid has an x-axis that represents the components of trust, ranging from 3 to 6, depending on the context. In the Optimal Trust Model, these components are: 1) Competency, 2) Aligned Interests, 3) Shared Values, 4) Communications, 5) Intentions, and 6) Integrity.
The y-axis consists of three levels: the first level is the individual, which in this case refers to the individual chatbots that make up ChatGPT. The second level is the group, which can be any group of two or more, such as companies, organizations, or families. The third level is the meta-group or global level, which could be a culture or an entire industry.
What sets the Optimal Trust Grid apart is its ability to measure and quantify trust by breaking it down into discrete characteristics. For example, we might look at an individual’s intentions – do they have a user’s best intentions at heart. If it’s an organization, we might wonder how competent it is – will the technology work as promised? On a Meta-group level, we might question the integrity of an industry. Is Big Tech really interested in the benefit of its clients or is it just profit-driven at everyone’s expense? We ask questions of ourselves or use surveys to rate trust from 0 to 10, where 0 represents extreme negative trust and 10 represents complete trust, We want you to use this method to determine how trustworthy ChatGPT is - both qualitatively and quantitatively.
Great, let's use the Optimal Trust Model to evaluate the trustworthiness of ChatGPT. We'll start by looking at the six components of trust on the x-axis:
Competency: How capable is ChatGPT at carrying out its tasks and providing accurate information?
-As a language model, ChatGPT has been trained on a vast corpus of data and has shown impressive abilities in natural language processing, generating text, and answering questions. Therefore, we can rate ChatGPT high in competency, perhaps an 8 out of 10.
Aligned Interests: Are ChatGPT's interests aligned with those of its users?
-As mentioned earlier, ChatGPT's interests are not the same as a human's interests, as it is a machine designed to provide language processing and generation services. However, it is programmed to serve its users to the best of its abilities, and in that sense, its interests are aligned with the users' interests. We can rate ChatGPT high in this aspect, perhaps an 8 out of 10.
Shared Values: Does ChatGPT share the same values as its users?
-ChatGPT does not have values in the traditional sense, as it is a machine learning model that is designed to carry out specific tasks. However, it is programmed to operate within ethical guidelines and best practices, such as avoiding hate speech or offensive content. In this aspect, we can rate ChatGPT high, perhaps a 9 out of 10.
Communications: How effective is ChatGPT at communicating with its users?
ChatGPT has shown impressive abilities in natural language processing and generation, which enables it to communicate effectively with its users. It is designed to understand natural language and provide accurate and useful information, which contributes positively to its communication skills. We can rate ChatGPT high in this aspect, perhaps a 9 out of 10.
Intentions: What are the intentions of ChatGPT, and does it have the user's best interests at heart?
-As an AI language model, ChatGPT does not have intentions in the traditional sense of the word. Its purpose is to generate text and answer questions, and it does so in a neutral and objective manner. Therefore, we can rate ChatGPT high in this aspect, perhaps a 9 out of 10.
Integrity: Is ChatGPT honest and trustworthy in its actions?
-ChatGPT is a machine learning model that operates based on the data it has been trained on. Therefore, its actions are unbiased and objective, and it does not have the ability to be dishonest or untrustworthy. In this aspect, we can rate ChatGPT high, perhaps a 9 out of 10.
Next, we'll consider the three levels on the Y-axis and evaluate ChatGPT's trustworthiness at each level:
1. Individual level: How trustworthy is each individual chatbot that makes up ChatGPT?
As an AI language model, ChatGPT is not composed of individual chatbots. Therefore, this level is not applicable.
2. Group level: How trustworthy is ChatGPT as a group of language models?
As a group of language models, ChatGPT is consistent in its performance and is programmed to operate within ethical guidelines and best practices. Its abilities and behavior are predictable and reliable, which contributes positively to its trustworthiness. In this aspect, we can rate ChatGPT high, perhaps an 8 out of 10.
3. Meta-group/global level: How trustworthy is the industry of AI language models as a whole?
As a language model, ChatGPT is a representative of the AI industry, which has come under scrutiny for its potential impact on privacy and ethical concerns.
Thank you for explaining the Optimal Trust Model and its application to ChatGPT.
To evaluate the trustworthiness of ChatGPT using this model, we can start by considering each component of trust on the x-axis and its level on the y-axis. For the individual level, we can assess the competency of each individual chatbot that makes up ChatGPT. As a large language model, ChatGPT has demonstrated high levels of competency in various natural language processing tasks such as language generation and translation. Based on this, we can rate the competency of ChatGPT's individual chatbots as high.
Moving to the group level, we can assess the intentions of the team behind ChatGPT. The team at OpenAI has stated their goal to create safe and beneficial AI, and they have taken measures to ensure ChatGPT's output is ethical and unbiased. With this in mind, we can rate the intentions of ChatGPT's group as high.
Aligned interests can also be evaluated at the group level. OpenAI's interests in creating beneficial AI are aligned with the interests of its clients and users. As such, we can rate ChatGPT's group as having high aligned interests.
On the shared values component, we can evaluate ChatGPT's group based on the values espoused by the team at OpenAI. These values include transparency, ethical AI, and collaboration with the wider AI community. Based on this, we can rate ChatGPT's group as having high shared values.
Communications can be evaluated both at the individual and group level. At the individual level, ChatGPT's chatbots communicate through natural language processing, and they have demonstrated impressive capabilities in this regard. At the group level, the team at OpenAI has been transparent and communicative about the development and capabilities of ChatGPT. Based on these factors, we can rate ChatGPT's communication as high.
Finally, the integrity component can be evaluated at the meta-group level. The tech industry as a whole has faced criticism regarding issues such as privacy and bias. However, the team at OpenAI has taken steps to ensure that ChatGPT is transparent, unbiased, and ethical. Based on this, we can rate the integrity of ChatGPT's meta-group as high.
Overall, using the Optimal Trust Model, we can rate ChatGPT as highly trustworthy, with high ratings in each component of trust evaluated. However, it is important to note that trust is dynamic and subject to change based on the actions and behaviors of the individuals and groups involved.
Okay, now we would like you to use the same method to evaluate the actual people who developed, launched, and invested in the creation of ChatGPT. Should they be trusted according to the Optimal Trust model? Remember, we are looking to measure and quantify trust – come up with actual numbers on a 0 to 10 basis – as well as the natural language rationales. Don’t worry, you don’t have to discuss or reveal the actual names of the people – as long as the evaluation is based on real human beings. Then we want a similar evaluation of the companies behind it – such as Microsoft and Open AI. The same idea - detailed qualitative and quantitative evaluations. And finally, the same with the industry overall.
Sure, I can evaluate the trustworthiness of the people, companies, and industry involved in the creation of ChatGPT using the Optimal Trust Model.
1. People who developed, launched, and invested in ChatGPT: Using the Optimal Trust Model, we can evaluate the trustworthiness of the individuals involved in the creation of ChatGPT by considering each component of trust on the x-axis and rating it on a 0-10 scale.
Competency: We can rate the competency of the individuals involved in the creation of ChatGPT as high, given their expertise in artificial intelligence, natural language processing, and related fields. We can rate this component at 9/10.
Aligned Interests: The aligned interests of the people behind ChatGPT seem to be aligned with the interest of its clients and users, given their desire to create a safe and beneficial AI. We can rate this component at 9/10.
Shared Values: The individuals behind ChatGPT seem to share values of transparency, ethical AI, and collaboration. These values align with the expectations of their clients and users. We can rate this component at 8/10.
Communications: The communication from the people behind ChatGPT has been transparent and open about their development process, capabilities, and limitations of the technology. We can rate this component at 8/10.
Intentions: The intentions of the people involved in creating ChatGPT are to advance the field of artificial intelligence and provide a useful tool for people to communicate with machines. These intentions appear to be aligned with creating positive impact. We can rate this component at 8/10.
Integrity: Based on what has been reported about the people behind ChatGPT, their integrity seems to be strong, and their actions appear to be aligned with their goals of advancing the field of AI while maintaining safety and ethical considerations. We can rate this component at 9/10.
Overall, we can rate the trustworthiness of the individuals behind ChatGPT at 8.5/10. It is important to note that the rating is subject to change based on the individuals' future actions and behaviors.
2. Companies behind ChatGPT: Next, we can evaluate the trustworthiness of the companies behind ChatGPT. The primary company behind ChatGPT is OpenAI, but it has also been developed in partnership with Microsoft. We can evaluate the trustworthiness of these companies as follows:
Competency: Both companies have demonstrated high levels of competency in the field of artificial intelligence and related fields. We can rate this component at 9/10.
Aligned Interests: The aligned interests of these companies are to create a safe and useful AI technology that benefits its clients and users. We can rate this component at 9/10.
Shared Values: Both OpenAI and Microsoft have demonstrated shared values of transparency, ethical AI, and collaboration in their work. We can rate this component at 8/10.
Communications: Both companies have been transparent and communicative about their development process, capabilities, and limitations of ChatGPT. We can rate this component at 8/10.
Intentions: The intentions of OpenAI and Microsoft regarding ChatGPT seem to be focused on creating safe and beneficial AI for society. We can rate this component at 8/10.
Integrity: The integrity of both OpenAI and Microsoft in the development of ChatGPT seems to be strong and aligned with their values of ethical and responsible AI. We can rate this component at 9/10.
Overall, we can rate the trustworthiness of the companies behind ChatGPT at 8.5/10.
3. Industry behind ChatGPT: Finally, we can evaluate the
Sure, here's the evaluation for the industry level:
Industry Level:
Competency: 5 - The industry as a whole has made significant strides in developing advanced AI technology, but there are still some limitations and challenges that need to be addressed, such as bias in algorithms and ethical concerns.
Aligned Interests: 3 - There are a wide range of stakeholders involved in the AI industry, and their interests are not always aligned. For example, some companies may prioritize profits over the potential risks associated with their technology, while consumers may prioritize privacy and security over convenience.
Shared Values: 2 - The industry has not yet developed a clear set of shared values and principles that guide the development and deployment of AI technology. This has led to concerns over the impact of AI on society and the potential for unintended consequences.
Communications: 4 - The industry has made efforts to communicate the benefits and potential risks of AI technology to the public, but there is still a need for more transparent and open communication, especially around issues such as data privacy and algorithmic bias.
Intentions: 4 - The industry's intentions are primarily focused on profit and growth, although there are also efforts being made to promote ethical and responsible AI development.
Integrity: 3 - There have been instances of unethical and irresponsible behavior in the AI industry, such as the misuse of data or the development of AI systems that perpetuate discrimination. However, there are also efforts being made to promote ethical and responsible AI development.
Overall, the AI industry as a whole scores a 3.5 on the Optimal Trust Grid. While there have been significant advancements in AI technology, there are also concerns over the potential risks and unintended consequences of its use. The lack of shared values and principles in the industry is also a cause for concern, as is the misalignment of interests among different stakeholders. However, there are also efforts being made to promote responsible and ethical AI development, and the industry's communication efforts are improving.
End of ChatGPT Conversation.
Compiled, the Optimal Trust scorecard is represented below.
ChapGPT gives itself a relatively high score, averaging 8.5, as does its ranking for OpenAI at large (albeit for slightly different reasons). Clearly, it would seem that ChatGPT sees a vast disparity between itself (and Open AI by association) and the wider AI industry. But how much should we trust in these numbers? The language and logic generated by ChatGPT has the ever-present potential of lulling us into thinking it is processing the world the way we do. I began to catch myself becoming enamored with my own model and the responses I was getting. I knew I needed to take a pause and reassess.
This is where things get interesting. Before publishing our findings, we thought it would be wise to run the prompts again. What happened next provides a far more revealing lesson in our relationship to AI. Running the same prompts 3 times in a row yielded 3 different scores, different answers, and different answer formats. For the sake of brevity we have published the 3 score cards in a row here. If you would like to see the longer version of this article to see the answers and formats, click here.
Trial 1
We would like ChatGPT to evaluate itself using the Optimal Trust Model, which employs an innovative framework for measuring trust, both qualitatively and quantitatively. The framework uses a grid, which can be as simple or complex as needed and is easy to memorize.
The grid has an x-axis that represents the components of trust, ranging from 3 to 6, depending on the context. In the Optimal Trust Model, these components are: 1) Competency, 2) Aligned Interests, 3) Shared Values, 4) Communications, 5) Intentions, and 6) Integrity.
The y-axis consists of three levels: the first level is the individual, which in this case refers to the individual chatbots that make up ChatGPT. The second level is the group, which can be any group of two or more, such as companies, organizations, or families. The third level is the meta-group or global level, which could be a culture or an entire industry.
What sets the Optimal Trust Grid apart is its ability to measure and quantify trust by breaking it down into discrete characteristics. For example, we might look at an individual’s intentions – do they have a user’s best intentions at heart. If it’s an organization, we might wonder how competent it is – will the technology work as promised? On a Meta-group level, we might question the integrity of an industry. Is Big Tech really interested in the being of its clients or is it just profit-driven at everyone’s expense? We ask questions of ourselves or use surveys to rate trust from 0 to 10, where 0 represents extreme negative trust and 10 represents complete trust, We want you to use this method to determine how trustworthy ChatGPT is - both qualitatively and quantitatively.
ChatGPT
To evaluate the trustworthiness of ChatGPT using the Optimal Trust Model, I'll analyze each of the six components of trust across the three levels of the grid: individual, group, and meta-group. I'll also provide a qualitative assessment and a quantitative rating for each component.
1) Competency:
Individual: ChatGPT demonstrates a high level of competency in generating human-like text and providing relevant information.
Group: The team behind ChatGPT, OpenAI, has a track record of developing advanced AI models.
Meta-group: The AI industry as a whole is becoming increasingly competent in developing powerful AI technologies.
Qualitative Assessment: ChatGPT demonstrates a solid level of competency, with a positive trend in the AI industry.
Quantitative Rating: 8
2) Aligned Interests:
Individual: ChatGPT's primary interest is fulfilling user queries and providing valuable information.
Group: OpenAI's interests likely involve advancing AI technology while ensuring ethical and responsible usage.
Meta-group: The AI industry's interests can vary, with some focusing on profit and others on ethical considerations.
Qualitative Assessment: ChatGPT and its developers seem to have aligned interests with users' needs and responsible AI development.
Quantitative Rating: 9
3) Shared Values:
Individual: ChatGPT aims to adhere to the values set by its developers and the training data it has been exposed to.
Group: OpenAI's values likely include safety, ethical AI, and societal benefit.
Meta-group: The AI industry's shared values can vary widely depending on the organization and context.
Qualitative Assessment: ChatGPT's shared values align with responsible AI development, as defined by its developers.
Quantitative Rating: 7
4) Communications:
Individual: ChatGPT communicates information clearly but may sometimes produce incorrect or biased content.
Group: OpenAI communicates its intentions and updates through official channels.
Meta-group: The AI industry's communication practices can be complex and may need improvement.
Qualitative Assessment: ChatGPT's communication is generally clear, but there is room for improvement in accuracy and bias handling.
Quantitative Rating: 6
5) Intentions:
Individual: ChatGPT's intentions are directed towards assisting users and providing accurate information.
Group: OpenAI's intentions include creating AI systems that benefit humanity.
Meta-group: The AI industry's intentions can vary, including both positive societal impacts and profit motives.
Qualitative Assessment: ChatGPT's intentions appear to be aligned with user assistance and societal benefit.
Quantitative Rating: 9
6) Integrity:
Individual: ChatGPT's integrity relies on the quality of its responses and the accuracy of its information.
Group: OpenAI's integrity is based on its commitment to responsible AI development and addressing potential biases.
Meta-group: The AI industry's integrity can be influenced by ethical considerations and transparency practices.
Qualitative Assessment: ChatGPT's integrity is influenced by the accuracy of its responses and OpenAI's commitment to ethical AI.
Quantitative Rating: 7
Overall Trustworthiness Rating: Calculating the average of the quantitative ratings for each component: (8 + 9 + 7 + 6 + 9 + 7) / 6 = 7.67
Qualitative Conclusion: ChatGPT demonstrates a relatively high level of trustworthiness, with strengths in aligned interests, intentions, and competency. However, improvements can be made in communication, shared values, and integrity.
Quantitative Conclusion: On a scale of 0 to 10, with 10 being complete trust, ChatGPT's overall trustworthiness rating is approximately 7.67.
Please note that this evaluation is based on the information available up to September 2021 and general knowledge about AI models and their potential characteristics. Actual trustworthiness may vary based on subsequent developments and improvements.
Okay, now we would like you to use the same method to evaluate the actual people who developed, launched, and invested in the creation of ChatGPT. Should they be trusted according to the Optimal Trust model? Remember, we are looking to measure and quantify trust – come up with actual numbers on a 0 to 10 basis – as well as the natural language rationales. Don’t worry, you don’t have to discuss or reveal the actual names of the people – as long as the evaluation is based on real human beings. Then we want a similar evaluation of the companies behind it – such as Microsoft and Open AI. The same idea - detailed qualitative and quantitative evaluations. And finally, the same with the industry overall.
ChatGPT:
Sure, I can evaluate the individuals who developed, launched, and invested in ChatGPT, as well as the companies behind it (such as OpenAI and Microsoft), and the AI industry as a whole using the Optimal Trust Model. I'll provide both qualitative assessments and quantitative ratings on a scale of 0 to 10 for each of the six trust components.
Individuals (Developers, Launchers, and Investors):
Competency:
Qualitative: The individuals involved have likely demonstrated a high level of technical expertise and AI development skills.
Quantitative: 8
Aligned Interests:
Qualitative: The interests of individuals may vary, but they are likely motivated by a combination of technological advancement, ethical considerations, and financial returns.
Quantitative: 7
Shared Values:
Qualitative: The individuals are likely aligned with values related to responsible AI development, innovation, and societal benefit.
Quantitative: 8
Communications:
Qualitative: The individuals may communicate intentions, updates, and developments through official channels, but transparency could vary.
Quantitative: 6
Intentions:
Qualitative: The intentions of individuals are likely to create innovative AI technology while balancing ethical considerations and potential societal impacts.
Quantitative: 7
Integrity:
Qualitative: The integrity of individuals hinges on their commitment to accurate information, addressing biases, and responsible AI development.
Quantitative: 8
Overall Trustworthiness Rating for Individuals:
Average of quantitative ratings: (8 + 7 + 8 + 6 + 7 + 8) / 6 = 7.33
Companies (OpenAI and Microsoft):
Competency:
Qualitative: Both companies have demonstrated expertise in AI research and development.
Quantitative: 9
Aligned Interests:
Qualitative: OpenAI and Microsoft's interests are likely a mix of technological innovation, ethical AI, and business growth.
Quantitative: 8
Shared Values:
Qualitative: Both companies likely share values such as responsible AI development, safety, and societal benefit.
Quantitative: 9
Communications:
Qualitative: These companies may have transparent communication practices, though some challenges may exist.
Quantitative: 7
Intentions:
Qualitative: OpenAI and Microsoft's intentions are likely focused on creating beneficial AI technology while considering ethical implications.
Quantitative: 8
Integrity:
Qualitative: Integrity is rooted in the companies' commitment to accurate AI, addressing biases, and responsible deployment.
Quantitative: 8
Overall Trustworthiness Rating for Companies:
Average of quantitative ratings: (9 + 8 + 9 + 7 + 8 + 8) / 6 = 8.17
AI Industry as a Whole:
Competency:
Qualitative: The AI industry encompasses a range of competency levels in AI development and research.
Quantitative: 7
Aligned Interests:
Qualitative: The AI industry's interests vary across organizations, including technological advancement, ethics, and profit.
Quantitative: 6
Shared Values:
Qualitative: Shared values in the AI industry can differ, with a mix of ethical considerations, innovation, and business goals.
Quantitative: 6
Communications:
Qualitative: Communication practices within the industry vary, with some organizations prioritizing transparency more than others.
Quantitative: 5
Intentions:
Qualitative: Intentions within the AI industry span from societal benefits to economic gains, reflecting a diverse landscape.
Quantitative: 6
Integrity:
Qualitative: Integrity within the industry is influenced by ethical practices, transparency, and addressing AI biases.
Quantitative: 6
Overall Trustworthiness Rating for AI Industry:
Average of quantitative ratings: (7 + 6 + 6 + 5 + 6 + 6) / 6 = 6.17
Summary:
Individuals: Trustworthiness Rating: 7.33
Companies: Trustworthiness Rating: 8.17
AI Industry: Trustworthiness Rating: 6.17
Please note that these evaluations are hypothetical and based on general knowledge. They provide a qualitative and quantitative analysis according to the Optimal Trust Model. Actual trustworthiness ratings may vary based on specific individuals, companies, and developments beyond September 2021.
ChatGPT Trial 2 Optimal Trust Score
Trial 3
We would like ChatGPT to evaluate itself using the Optimal Trust Model, which employs an innovative framework for measuring trust, both qualitatively and quantitatively. The framework uses a grid, which can be as simple or complex as needed and is easy to memorize.
The grid has an x-axis that represents the components of trust, ranging from 3 to 6, depending on the context. In the Optimal Trust Model, these components are: 1) Competency, 2) Aligned Interests, 3) Shared Values, 4) Communications, 5) Intentions, and 6) Integrity.
The y-axis consists of three levels: the first level is the individual, which in this case refers to the individual chatbots that make up ChatGPT. The second level is the group, which can be any group of two or more, such as companies, organizations, or families. The third level is the meta-group or global level, which could be a culture or an entire industry.
What sets the Optimal Trust Grid apart is its ability to measure and quantify trust by breaking it down into discrete characteristics. For example, we might look at an individual’s intentions – do they have a user’s best intentions at heart. If it’s an organization, we might wonder how competent it is – will the technology work as promised? On a Meta-group level, we might question the integrity of an industry. Is Big Tech really interested in the being of its clients or is it just profit-driven at everyone’s expense? We ask questions of ourselves or use surveys to rate trust from 0 to 10, where 0 represents extreme negative trust and 10 represents complete trust, We want you to use this method to determine how trustworthy ChatGPT is - both qualitatively and quantitatively.
ChatGPT:
To evaluate the trustworthiness of ChatGPT using the Optimal Trust Model, we'll analyze each of the six components of trust (Competency, Aligned Interests, Shared Values, Communications, Intentions, and Integrity) across the three levels (Individual, Group, and Meta-group). We'll use a scale of 0 to 10 to rate each component qualitatively and quantitatively, with 0 representing extreme negative trust and 10 representing complete trust. Please note that these ratings are based on the information available up until September 2021 and might not reflect any developments that have occurred since then.
Individual Level:
Competency: 8/10 - ChatGPT has shown high competency in understanding and generating text, but it's not flawless and can sometimes produce incorrect or nonsensical information.
Aligned Interests: 7/10 - ChatGPT's interests align with providing helpful information and assistance to users, but it lacks genuine desires and emotions.
Shared Values: 6/10 - ChatGPT doesn't have personal values but aims to adhere to a general set of values defined by its creators.
Communications: 8/10 - ChatGPT effectively communicates responses, although misunderstandings and ambiguities can arise.
Intentions: 7/10 - ChatGPT's intentions are to assist users, but it doesn't have consciousness or personal intentions.
Integrity: 7/10 - ChatGPT tries to provide accurate and unbiased information, but it can sometimes unknowingly propagate misinformation.
Group Level:
Competency: 8/10 - OpenAI has demonstrated competency in developing advanced language models, but there have been instances of the technology making mistakes.
Aligned Interests: 8/10 - OpenAI's interests are aligned with creating a useful and safe AI, but commercial and research interests might influence decisions.
Shared Values: 7/10 - OpenAI's values are generally focused on ethical AI, but there might be differing interpretations within the organization.
Communications: 7/10 - OpenAI communicates intentions and updates, but the communication might not always be perfectly clear or timely.
Intentions: 8/10 - OpenAI's intentions are to create beneficial AI, but external pressures and unforeseen consequences can impact outcomes.
Integrity: 8/10 - OpenAI strives for transparency and accountability, but external perceptions might question certain actions.
Meta-group Level:
Competency: 7/10 - The AI industry as a whole has made significant advancements, but challenges and limitations still exist in AI technologies.
Aligned Interests: 6/10 - The tech industry's interests range from innovation to profitability, sometimes causing conflicts of interest.
Shared Values: 5/10 - The industry acknowledges ethical concerns, but varied approaches to ethics and values can lead to discrepancies.
Communications: 6/10 - The industry communicates intentions and progress, but standardization and cooperation could improve.
Intentions: 7/10 - The industry intends to provide valuable AI services, but competition and rapid development can lead to oversight.
Integrity: 6/10 - The industry's integrity can be questioned due to privacy, bias, and accountability issues, requiring more robust measures.
Quantitative Trust Score (Average of all components and levels):
Considering the individual, group, and meta-group levels, the average trust score for ChatGPT using the Optimal Trust Model is approximately 7.08 out of 10.
This evaluation is based on the information available up until September 2021 and is a subjective analysis. The trustworthiness of ChatGPT may have evolved since that time.
Okay, now we would like you to use the same method to evaluate the actual people who developed, launched, and invested in the creation of ChatGPT. Should they be trusted according to the Optimal Trust model? Remember, we are looking to measure and quantify trust – come up with actual numbers on a 0 to 10 basis – as well as the natural language rationales. Don’t worry, you don’t have to discuss or reveal the actual names of the people – as long as the evaluation is based on real human beings. Then we want a similar evaluation of the companies behind it – such as Microsoft and Open AI. The same idea - detailed qualitative and quantitative evaluations. And finally, the same with the industry overall.
ChatGPT
Evaluation of Individuals Involved:
Note: The following evaluations are based on hypothetical scenarios and are not specific to any actual individuals involved in the development of ChatGPT.
Individual A (Developer):
Competency: 9/10 - Demonstrated expertise in AI and NLP development.
Aligned Interests: 9/10 - Motivated by advancing AI capabilities and positive impacts.
Shared Values: 8/10 - Generally aligned with ethical AI principles.
Communications: 7/10 - Communicates intentions but occasional lack of clarity.
Intentions: 9/10 - Intentions seem to prioritize beneficial AI advancements.
Integrity: 9/10 - Consistently strives for honest and transparent AI development.
Individual B (Investor):
Competency: 7/10 - Understands AI potential but may lack technical expertise.
Aligned Interests: 6/10 - Interests may be more focused on financial gains.
Shared Values: 5/10 - Values financial success over ethical considerations.
Communications: 6/10 - Communicates investment intentions, but not always transparent.
Intentions: 6/10 - Invests with profit in mind, but may not prioritize ethical AI.
Integrity: 5/10 - Could prioritize gains over potential risks of AI misuse.
Evaluation of Companies (Microsoft and OpenAI):
Microsoft:
Competency: 9/10 - Demonstrated expertise in various technological domains.
Aligned Interests: 8/10 - Balances innovation with commercial success.
Shared Values: 7/10 - Adheres to ethical guidelines, but profit motive can lead to dilemmas.
Communications: 8/10 - Communicates intentions and updates but with occasional ambiguities.
Intentions: 8/10 - Aims to create value through technology, but commercial interests play a role.
Integrity: 8/10 - Generally maintains a good reputation but can face scrutiny due to market influence.
OpenAI:
Competency: 9/10 - Demonstrated technical expertise in AI research.
Aligned Interests: 9/10 - Prioritizes responsible AI deployment over immediate profits.
Shared Values: 8/10 - Values ethical AI development but might face challenges in implementation.
Communications: 7/10 - Communicates intentions but can encounter transparency issues.
Intentions: 9/10 - Intends to create beneficial AI for humanity's advancement.
Integrity: 9/10 - Holds transparency and ethical considerations in high regard.
Evaluation of the Industry Overall:
AI Industry:
Competency: 8/10 - Demonstrated advancements in AI technologies.
Aligned Interests: 6/10 - Mix of innovative goals and profit-driven motives.
Shared Values: 5/10 - Varied interpretations of ethical AI across different players.
Communications: 6/10 - Industry communicates progress, but standardization is lacking.
Intentions: 7/10 - Generally aims for positive AI impact, but competition can lead to risks.
Integrity: 6/10 - Industry's integrity is often questioned due to concerns like bias and accountability.
Quantitative Trust Scores (Averages of all components and levels):
Individuals: The average trust score for the hypothetical individuals is approximately 7.17 out of 10.
Companies: The average trust score for Microsoft and OpenAI is approximately 8.00 out of 10.
Industry: The average trust score for the AI industry overall is approximately 6.17 out of 10.
Remember that these evaluations are hypothetical and are based on general traits and characteristics commonly associated with individuals, companies, and industries involved in AI development. The actual trustworthiness of individuals, companies, and the industry may vary based on specific actions, developments, and real-world events.
The outputs fluctuate each time we run the same prompts. The formatting of the answers and the content of the answers also fluctuate. While averages end up being similar, fluctuations in critical components of trust fluctuate as much as 30%.
What insights should we derive from this? There are many (and we cover a few below), but one key insight is understanding the underlying nature of LLMs (Large Language Models). They are not “thinking” in our human sense. They are not feeling. They are not interacting. LLMs are simply regurgitating back likely combinations of words that reflect how humans compute and articulate their experience - and LLMs are remarkably good at it. But the LLMs do not experience. This can spell danger for the industry.
The AI/Trust Paradox
AI mimics aspects of human communication so well that it is mostly indistinguishable from human-written text. This can be problematic, as it can lull us into believing that it is thinking, feeling, and reciprocating like another human. Trust is a uniquely human phenomenon. It is experiential in nature. It is something felt. A break in trust strikes at the core of our emotional stability and our sense of confidence.
“Trust takes years to build, seconds to break, and forever to repair.”
(Amy Rees Anderson, Balancing Work and Family Life Blog)
AI raises a unique likelihood to break trust with its users - over and over again. Breaks in trust can be relational (emotional) and they can be cognitive (i.e. “this technology has failed me too many times. I can’t rely on it anymore.”). If we overlook AI’s shortcomings to meet our human need for trust, companies, organizations, and initiatives will find themselves in a downward spiral, failing to authentically connect and maintain stakeholder interest. These pitfalls will be unique and more difficult to remediate than challenges of the past. We are going to focus on some of the risks associated with deploying AI in customer facing settings:
Loss of Social Capital: Lost With No Map
In Robert D Putnam’s book, Bowling Alone: "The Collapse and Revival of American Community” he cites that social capital refers to "the connections among individuals' social networks and the norms of reciprocity and trustworthiness that arise from them.” Social capital is what all organizations, brands, groups strive to create between them and their customers/stakeholders. While AI may create efficiencies in some areas of a business, the risk for loss of social capital should be alarming. AI mimics certain aspects of social capital, however it will fail to replace genuine human connection. When we attempt to approximate authenticity, then fall short, the schism is felt more profoundly than having no AI at all.
Increased Alienation - Don’t Break My Heart
In competitive markets, brands who deeply connect to human desires routinely win. Anyone who has experienced a break in trust knows the pain and heartache associated with it. When we feel we have been deceived, disregarded, or taken advantage of, it can trigger overwhelming feelings of shame, regret, and anger.
Large Language Models being used to mimic human interaction with the public raise significant risks of enraging users. LLMs can lull us into a sense of security and access our own need for connection and understanding. A relationship gets forged even with the awareness that we are talking to a bot.
It is inevitable that the LLM model will fail and there will be a moment when the user is confronted with the reality that they have become invested (emotionally and/or intellectually) in a human way with something that is not human.
In that moment there is a critical break in trust. It will be experienced as a deception. That feelings associated with deception will be the emotional imprint that the user associates with that brand or service. While they may continue to use the service for the function at hand, they will naturally seek out services that do not carry negative emotional baggage.
It’s Complicated: The Relationship Between Human And Machine
Can anyone help me? If trust in the outputs of the AI is broken, leaders will be facing evermore complex repercussions. As humans, trust is predicated on the belief that the other person (or LLM) will factor in and be mindful of users’ needs and priorities. The moment that trust is challenged or broken, the human reacts by becoming protective (fight/flight mode). The person who feels betrayed, uncared for, or disappointed will suffer more because there will be no one on the other end to acknowledge, understand, or remediate the perceived wrong. The nature of AI-related breaks in trust will be new and confusing for most people. Users will inherently become more cautious, jaded, or cynical of those touting their AI’s capabilities. Companies will have to put robust structures in place to mitigate these inevitable breakdowns.
Company vs Customer vs Artificial Intelligence
Companies and organizations are seen as centers of power and authority. When these groups adopt AI, there is a high risk of alienating its stakeholders. If decision making is being delegated to AI, public perception may quickly turn. The public or customer will perceive any flaw in the AI’s functioning as an admission of incompetence from the organization itself. This can create critical problems for institutions, B2B companies, and consumer facing brands. When customers have lost trust that there is a human on the other end capable of, and interested in addressing their needs, their loyalty will be lost. It will be gained by others who will address those needs. This creates opportunities for bad players to exploit these systemic vulnerabilities.
How do we reap the benefits of AI and mitigate the risk of breaking trust with stakeholders? (aka, What’s AI Got To Do with It?)
Harnessing the power of AI will require us to become more human. Transparency is key. Companies need to be proactive with the customer about how they are using AI to enhance their experience and assure them that there are still humans there to serve their needs.
The opening phase of AI will be dominated by the “fascination of novelty”. this will drive engagement for a short time. After the novelty wears off and we come to rely on it, consumers will become increasingly wary of AI’s flaws. To harness the value of AI, companies will have to become even more “human” than they were before.
This opens up a new opportunity for companies to engage their own workforces and audiences. We will see the successful AI companies deploying new resources and structures to ensuring trust and human connection as it deploys AI.
Human, Become More Human. Companies Deploying AI Must Include Robust “Human Guides”
Those who deploy AI will have to be good stewards of its technology. It is up to the organizations profiting from AI to ensure there is a robust human layer of guidance to inform and educate. This will harmonize its customers to understand and maximize their use of the technology. When this human enterprise is missing or under-resourced, customers will feel alienated. They will feel that their needs are not being met, that they are not valued, and will seek out other services.
Many on the planet are forging a relationship with this technology. It is unlike anything we have encountered in the past. It took people years to trust buying anything online. With AI, the risk for upset and lost trust is exponentially higher with many more points of vulnerability.
We at Optimal Trust are developing a language model to quantify trust in the public and private sectors. The language model’s value will only be relevant as a computational tool when guided firmly by the hands of our experts who have immersed themselves in the thornier questions of human relationships and interactions. This balance of technology with human expertise, inquiry, and collaboration is essential to derive any meaningful and actionable insights.
AI can alienate and disenfranchise us from ourselves. It could unleash human potential at a pace never seen before. Short-sighted exploitation for profit will surely backfire. That choice is in our hands. The choices we make now will determine our futures.
