Receive Free Artificial Intelligence Updates
I will send myFT Daily Digest E-mail summarizing the latest information artificial intelligence News every morning.
Igor Tulchinsky is the founder, chairman and CEO of WorldQuant, a global quantitative wealth management firm.
With today’s dizzying pace of innovation, it can be difficult to know where to focus your attention. This month alone saw news of Apple’s Vision Pro AR launch and DeepMind’s AlphaDev, an AI system for faster algorithmic identification. character. AI debuted a mobile chatbot app, and Intel and AMD both unveiled prototype chips. All of this rests on the potential for revelatory insights gained from continuous interaction with generative technologies and AI tools that can sometimes seem superhuman.
One way to maximize the benefits of the technology on offer is to approach the world of AI a little differently. Instead of asking what new things can be done with these tools, we should ask how they can solve problems that were previously thought unsolvable. Big leaps come not just from technological innovation, but from evolving perspectives on how to get the most out of these technologies.
Looking through my book Age of predictionIn a book I wrote with geneticist Professor Christopher Mason, I spent a lot of time looking at different prediction techniques. And you can’t enter the world of predictions unless you soon run into psychologist Philip Tetlock and his super-predictors.
Tetlock recognized a powerful truth many years ago. It is the wisdom of the wise crowd that is better than the wisdom of the crowd. He recruited hundreds of super-forecasters from academia, policy and business, and importantly recognized that half the art of good forecasting is asking the right questions.
Tetlock posed questions to super forecasters about everything from election results to stock market movements, asking them to assign probabilities to a preselected range of outcomes. The results were astonishing. His 2015 study by the U.S. Office of the Director of National Intelligence (ODNI) found that the super forecasters had all other forecasting methods and It turned out to be above technology.
However, there is one area where super forecasters tend to struggle. It’s a geopolitical negotiation. Anything that relies on the decisions of a few actors or the whims of an unpredictable leader is fraught with difficulty. They found it very difficult to predict, for example, the outcome of negotiations over the Iran nuclear deal or the behavior of the North Korean state under Kim Jong-un.
I believe AI can help address this weakness in the super forecaster’s skill set. Michael Milken once said that with enough research, anything can be predicted. The advantage of generative technology is that it requires algorithms to parse more data faster than humans were once able to, and allows them to test hypotheses more rigorously and effectively, thus making the lessons of hyper-prediction profound. It is applicable to scale. best ever.
When it comes to forecasting, ChatGPT is still hampered by the September 2021 data cut-off date, but companies like Meta and Google are making new large-scale investments that can better use today’s events to predict tomorrow’s outcomes. We have already built a language model for Combined with human intelligence, which dictates both the questions that are asked and how potential answers are presented, leads to what I believe to be the best of all possible worlds.
What does this mean? It’s clear that AI is streamlining and complementing super forecasters’ human thinking, rather than replacing it. Humans are still needed to make strategic and operational decisions. I believe that the role of AI is to improve the accuracy of that judgment. Human ingenuity is also required in prioritizing certain information.
Although AI can make rudimentary attempts to determine the credibility of sources, it often finds what appears to be counter-hallucinations in distinguishing between trustworthy sources and those that should be discarded. I have. Just as hallucinations in large-scale language models stem from some kind of eagerness to please, generative AI currently takes information at face value, for example, inventing supporting data when it doesn’t exist. They are so receptive that they seem unaware (or unwilling to consider) the information. There may be bad actors playing the information game, or the available sources the information game draws on are simply wrong or irrelevant.
Karl Popper, George Soros’ philosophical mentor, worked extensively on what he called “falsification theory.” It is an approach to information that employs a rigorous interpretation of the scientific method and provides a framework within which expected outcomes can be judged. According to Popper, science is a machine that destroys falsehoods, and every attempt to back up a thesis must begin with an attempt to disprove it. Understanding that AI is an imperfect tool when it comes to making predictions also helps us know when we can confidently claim that human thinking is superior to machine processing.
Claims can now be double tested. That means asking both the AI and my own group of super-predictors (at WorldQuant, through methods such as surveys sent to my colleagues) to refute my theories. It’s still imperfect, but it’s the least imperfect method I’ve employed so far when it comes to making a credible statement about the future of our rapidly changing world. I believe that the hyper-predictive example contains a broader truth. In short, the best applications of AI are those that seek to enhance and extend human ingenuity, not to replace it.
