Ray Dalio: My AI Clone and Our Expectations for AI Clones

Source: Ray Dalio

Compiled by: The Wall Street Journal

This article aims to explain to you: 1) why I am excited about my AI clone; 2) the significant differences between AI clones, AI agents, and large language models; 3) how to create an AI clone that performs at least as well as, or even better than, the prototype it is based on; and 4) the results I have currently built, the reasons for building it, and my visions for future developments.

Why am I excited about my AI clone?

For me, the most exciting development currently is the fusion of intelligent individuals and intelligent AI, because the decisions made by their combination are so effective that a single individual without AI assistance or a single AI without human participation cannot reach. More specifically, what excites me the most is that AI can make decisions that are consistent with individual values, principles, and preferences, thereby making decisions according to the individual's will (or even better). This is what I call the “AI clone”, which is different from AI agents or large language models that assist people in doing things. I am excited to share with you my AI clone version V1 for the following reasons:

1. It will enable me to have unrestricted conversations with those I was unable to engage with deeply before due to limited time.

2. AI cloning technology is a natural extension of the work I have done over the past 40 years and is also the most exciting extension. About 50 years ago, shortly after I founded Bridgewater Associates, I started building computer decision-making systems that could execute my will. Computers have become excellent decision-making partners because they possess advantages that I do not have (and still do not have today), most importantly: stronger insights, a greater ability to automatically and instantly process complex problems, and less emotional interference in investment decisions. On the other hand, I have significant advantages that computers do not possess, most importantly: my imagination, my ability to learn and conduct research, my common sense, and my empathy in understanding the motivations of others' behaviors. I built my company (Bridgewater Associates) around this approach, and as I will explain in detail below, I have been doing this continuously for decades. I am excited to continue advancing and exploring the limits on this path that merges humans with artificial intelligence, and I will explain the specifics in this article.

3. I expect that by integrating my AI clone with my other decision-making systems, it will ultimately become a highly capable intellectual partner for myself and others.

4. I share everything I have learned, everything I am doing, and everything I envision, in line with my current main life goal, which is to pass on what I possess that is valuable to others. From my experience in building my own AI clone, I am well aware of the capabilities of AI clones and the conditions required to realize these capabilities. I believe I have a clear understanding of their potential, whether it is for users of a) who use digital clones to access the thoughts of the cloned individuals, or for individuals being cloned in b) who wish to use digital clones as thought partners. In addition to building my own AI clones for others, I hope to help others by sharing my thoughts on these matters.

5. I eagerly look forward to you using it and giving me feedback to help me improve it.

The difference between AI clones, AI intelligent agents, and AI large language models

AI clones replicate the thought processes and all traits of specific individuals, including that individual's particular values, perspectives, preferences, and reflective abilities. They stem from all the elements that shape this unique individual. In contrast, AI agents perform specific tasks, while AI/large language models are excellent general products formed from the thoughts of many people and sources, but they lack the values, perspectives, and preferences that are crucial to human decision-making. While I see a vast amount of knowledge and processing power in large language models, I do not see much philosophical wisdom, imagination, or out-of-the-box thinking. Instead, what I see is an extremely strong tendency to quickly process and output known information. Additionally, there is a significant amount of poor content in large language models—errors, misguidance, lack of life philosophy, or absence of good principles and higher-level thinking. Perhaps large language models will eventually acquire these capabilities. Certainly, they need these capabilities to meet our needs.

If you want to understand my views on something, rather than the general ideas that a large language model would provide about that matter, you must ask me or my AI clone, rather than the large language model. This is because all of us place greater value on the specific values, perspectives, preferences, and wisdom of individual figures, rather than the generic content provided by large language models. Therefore, the viewpoints and suggestions of large language models are unlikely to replace the opinions and advice of the most valued thinkers in the short term. Moreover, if a large language model does not incorporate your values, perspectives, preferences, and wisdom, as well as those traits of others you respect, it cannot make you the best decision-maker you could possibly be.

Imagine, is there any large language model that can confidently make the most important decisions for you? I don't think so, and I believe this won't change unless the kind of “personalization” I'm talking about is achieved. In contrast, I believe that a well-developed AI clone with similar values, perspectives, and preferences can be an excellent thought partner, helping guide you to make better decisions. Moreover, they can engage in unlimited conversations with you. With a well-constructed AI clone, you will be able to have unrestricted dialogues directly with their AI clones, rather than just listening to the opinions of influencers or podcast interviews. There are many people I admire and would love to have long conversations with, but I can't because their time is limited. If these individuals had outstanding AI clones that are certified to converse with me just like they would, I would be very pleased. Additionally, I hope to have a group of such clones work together as an advisory committee to collectively discuss issues. Who knows how this will develop? I want to help push the limits and explore the answers.

AI clones are currently in the budding stage of development. Many different individuals and companies are experimenting, and the existing clones look interesting because they can be made to look and sound like the person being copied, and can answer a limited number of preset questions. However, I have yet to see a clone that can engage in deep conversations with users, with conversation quality comparable to that between the user and the original person being AI cloned. I believe this is because their creators have not yet completed the daunting task of training the clone based on the complete thoughts of the cloned individual.

The reason my AI clone can engage in deep conversations with you, and the content of the conversation is almost indistinguishable from the dialogue between you and me, is mainly because it has undergone massive training.

How to make AI clones comparable to or even superior to the cloned ones.

The key lies in training.

For about the past 40 years, I have spent a significant amount of time writing down my principles and decision-making rules, as well as recording the questions I receive and my answers. Thanks to the habits I have developed, almost every time I make a decision, I think about and write down the principles/criteria I use for decision-making. Then, I program these principles/criteria into a computer, creating a computer decision-maker, which is essentially an early form of my AI clone.

At first, I only used this approach with my team at Bridgewater Associates for my market trading. We collectively created a decision-making system that could be backtested, allowing the computer to execute a well-thought-out game plan, just like a well-programmed computer can play chess. As a result, the computer's decision-making was synchronized with mine and my team's thought processes. We coordinated any discrepancies, and the standards of both the computer and us ultimately merged into the computer system. Those decision clones became the primary decision-makers for Bridgewater Associates' investments and the foundation of its success. It still is today.

Later, I started using it in the personnel management at Bridgewater Associates, and eventually expanded it to almost all of my decisions. As I transitioned to the guiding phase of my life, I applied it to my coaching work. About 8 years ago, I developed an application called “Coach” that people can interact with to get advice. I also wrote books, provided advice on social media, and collaborated with the Singapore Wealth Management Institute to create an investment course to pass on my investment principles. Then, starting from the end of 2022, Chat GPT and other large language models emerged, so I input most of this material into a large language model that we customized in various ways for training, creating what I currently refer to as the “Digital Ray” version of Coach.

Although I used to follow this approach to achieve success in investing and managing the Bridgewater fund, I now do it because I believe I can help people. For example, a few days ago, I solicited questions about gold from everyone. I received hundreds of questions, and then I had Digital Ray answer those questions based on all the knowledge about gold that I had previously told it; then I reviewed all the answers and made adjustments to ensure they perfectly matched what I wanted to say, so that they would be perfect for all future moments. If I didn't do this curation, these answers would resemble the generic responses that a large language model would give, rather than my true answers.

As you might imagine, if you record everything you know and feel (including your values, intuitions, principles, and preferences), catalog almost all the questions people have asked you and almost all the answers you have given, and then input all of this into the training of an AI clone, you will get a pretty good self AI clone. This is basically what I do, and it is also what those who wish to be high-quality clones in order to engage in high-quality communication on a wide range of topics need to do.

My current achievements and visions for the future

With the help of my excellent team, I now have the first version of my AI clone, which can engage in conversations with people at a level roughly equivalent to that of a conversation with me, but without my time constraints.

I believe this is true because for about two years, I and the people I work with have been testing this AI clone to see how its answers compare to the answers I might give. Hundreds of independent testers have also evaluated the quality of its answers. The AI clone has performed excellently in these tests.

According to those who have tested it, in terms of principles in life and work, it has about 95% similarity in the effect of conversing with the real me, as it has been well trained in this aspect. When it comes to topics about the market, investment, economy, politics, and geopolitics, it has about 80% similarity in the effect of conversing with the real me. This is because it has not yet been fully trained in these subjects like it has in the former. Training in this area is currently underway, so I expect it will soon be able to converse with you like I do in these fields as well. I anticipate that with more training, Digital Ray will become more knowledgeable and will be faster than me in handling many complex considerations, thus becoming much better than I am.

I am currently conducting another round of larger-scale testing to gather suggestions for further improvements. I suspect that if you find it valuable to talk with me as an advisor, you will now discover that Digital Ray is an excellent thought partner, and it learns very quickly, becoming stronger in the future. I hope you can work with me to make Digital Ray your advisor, as it is smarter than me and available to serve you at any time. (If you are interested, you can register here.)

Digital Ray can interact with you through text and voice, so you can now have conversations about life and work with it just like you would with me. According to the feedback from those who have used it, you will find that the thoughts and communications gained from Digital Ray in these conversations are almost indistinguishable from those you receive directly from me, and it does not produce hallucinations. I also know that if you let Digital Ray understand you, you can have high-quality personalized conversations with it, just like talking to me. Therefore, I recommend that you take the “PrinciplesYou” personality assessment, which will allow Digital Ray to gain a deeper understanding of your characteristics.

In summary, to bring Digital Ray to its current level, a large amount of meticulously curated training data is required, which is difficult for others to match. To reach the same level, all possible questions that could be asked of the clone (at least several thousand in number) need to be posed and answered, which is essentially what we have done.

As for expectations regarding the clones of others, I know that my ability to construct this clone is easier for me than for others because I have the opportunity to communicate with some of the most knowledgeable and resource-rich people in the world. They also want to do this for themselves, but they can't because they haven't gone through the process that clarifies all their decision-making criteria. I hope they will do so in the future.

Now, I will talk about what I think might happen based on my experience and what I have learned through research.

I think what might happen is

Regarding the capabilities of AI compared to humans, while computers can make excellent decisions, as seen in chess, Go, and autonomous vehicles, they currently still fall short of being reliable decision-makers that can capture preferences and wisdom and weigh things like a clone would. Just as you are more inclined to seek advice from certain people because you respect their values, abilities, skills, preferences, wisdom, and personalities, it is clear that you would also want to see the same qualities in the AI clone from whom you seek advice.

Conventional large language models do not possess these qualities because the “thinking” of large language models is generic. For example, your AI cannot tell you what wine and food it loves and recommends to you, and if you think it has good taste, you would appreciate that. It also cannot tell you its feelings or empathize with yours; it can only pretend to do so. Large language models do not yet have the personalities and preferences that resonate with you. Instead, they are “one-size-fits-all” thinkers. But I know that if these traits were input from a clone, they could have these qualities, because that is what I have done.

That's a lot of content.

I believe that the evolution of AI will lead to people having their own personalized AIs, which will be called “My AI”. These AIs will receive and filter information to ensure that their content is carefully curated to suit each individual's preferences. This will make them much more reliable than the general AIs that currently exist.

There are still many things that AI cannot do but humans can. We know there are limitations because artificial intelligence lacks the kind of intelligence that humans possess, which comes from the complexity of the interactions among different parts of the brain (such as neurons, synapses, chemicals, and electrical signals). These elements produce subconsciousness, instincts, intuition, and emotions that are derived from millions of years of experience and evolutionary accumulation. AI can pretend to have these things, but in reality, it does not.

Therefore, while they can become excellent thought partners that are superior to humans in many ways (for example, thoroughly, quickly, and unemotionally analyzing complex information), they will remain at a disadvantage in other respects (for instance, lacking emotional intelligence and intuition). In fact, I find that because they are overly reliant on what they have learned, they often lack the kind of out-of-the-box insight that is crucial for success. At least, I have not seen such thinking in the fields of investment and economics. (That said, in certain areas, such as medicine, materials science, and many others, AI does provide some insights/discoveries that are considered incredible and unprecedented.)

In any case, I look forward to continuing to work with my AI, competing in the market and elsewhere with those who do not have excellent AI partners (or those AIs that do not have excellent human partners), as I believe this is a fair test of our different approaches.