Why do we need Hybrid Intelligence?
Original article: http://firrma.ru/data/articles/12610/?sphrase_id=40285
We are all aware of what artificial intelligence is, but what is hybrid intelligence? What is it needed for?
Hybrid Intelligence is the combination of human intelligence with machine intelligence and their interaction in resolving various tasks. One sort of intelligence supplements and strengthens the other one.
The uses of this type of intelligence and development of related systems are determined by the complexity of challenges. It is necessary when the data of a problem is largely indefinite. For example, venture investment is a task with a huge number of unknowns, so it is impossible to calculate the final variant.
In general, the history of this issue originated from a very simple task: How many jelly beans are their in the jar?
Finance professor Jack Treynor ran an experiment with a jar that held 850 beans of different colors. Upon the survey of all members of the class and calculation of the arithmetic average reply, he revealed the awesome accuracy of group answer, which was 871 beans.
Only one of the fifty-six people in the class made a better guess than the group average weighted result.
Upon the analysis of the results of many similar experiments, Jack Treynor made two important conclusions:
- Members of the group were not discussing the possible solutions and were working based on their individual guesses. This method provides highly accurate average results, provided there is a large sample.
— The group’s guess is sometimes not better than that of the most accurate members. In many (perhaps most) cases, there will be a few people who do better than the group’s mind. This is particularly important, when there is a selfish incentive for doing well.
A perfect example is the stock exchange, and we decided to apply this approach to it. The stock market is a true forecast. Today there is a share price and a number of unknowns, which will influence the price tomorrow. The market is very fast and needs numerous forecasts every day. Where should one get them? Certainly, from the analysts, but the number of experts is limited and all of them use basically the same information. Thus, in turn, often results in the ‘group thinking error’, when wrong decisions are made based on a single misleading insight. Incidentally, this is what happens on venture market.
We decided to run the beans-in-the-jar experiment on the stock exchange in order to test its efficiency. We asked users about the next day’s prices for different shares and tried to make deals based on their group opinion.
The experiments showed interesting results: In back tests, some 80% strategies showed annual yield over 20%, while our own account had a gain of 170% within six months. We decided to move further. We started to increase the sampling of analysts from a couple of dozens to a couple of thousands, added traders from algorithmic funds to the team and started to seek ways to both use the advantages of artificial intelligence and negate its disadvantages, namely bias and human emotional nature, which prevent people from showing stable accuracy results during a long period of time. Sure, group thinking did well, but we had a need for artificial intelligence.
We focused on the market and found the most successful and promising technological teams to resolve these tasks.
The New-York-based start-up Estimize, which raised a total of over $12 million in venture capital investment from the top fintech VC in New York, has been for five years beating the consensus of Wall Street analysts by predicting more accurate yields of public companies in their reporting periods.
The other startup, from the Western coast, is Number.ai. They raised $6 two months ago. In fact, they also collect group intelligence, but it’s machine group intelligence rather than human one. The company aggregates thousands of various trade algorithms and strategies in-house, gets them through self-learning artificial intelligence and therefore makes numerous deals on the exchange. In this case, our peers decided to totally abandon the human factor and rely on robots to make the decisions. However, this solution has a sound disadvantage: such systems are incapable of being rapidly adjusted for dramatic market changes, and each algorithm has a short lifetime, upon which it runs out.
Upon case study, we decided to align analysts and robot traders’ group intelligence and set up Cindicator. The synergy of the two types of intelligence is referred to as hybrid intelligence by scientists. This symbiosis allows for efficient resolution of the problems in both systems separately:
- group thinking error due to biased ideas of a narrow group of analysts and their centralisation;
- the absence of real-time system adaptation to any external (market) changes: crises, adjustments, unexpected news etc).
Keeping in mind the simple beans-in-the-jar experiment, we decided to run something similar and did so in partnership with Moscow Exchange.
During three weeks, 1,000 people predict the next day’s prices for four futures on a daily basis, while our Al robot trader models deals with the use of both averagely weighted answer of the crowd and our trading and mathematical algorithms.
In less than three weeks, the investment portfolio of this group has already increased by 3%, which accounts for 47% p.a. Let’s compare with the return of the world’s top funds:
CQS Directional Opportunities — 30%
Mudrick Distressed Opportunity Fund: 35.5%
Proxima Capital LP: 44,2%
Most funds are in the 12%-16% range.
Certainly, we understand that they have a much longer track record, but we look quite attractive against these figures.