“Quantity has a quality all its own” is a quote often attributed to Joseph Stalin. Recent results in AI show how this works for thinking.
Here is a short video from OpenAI. OpenAI is a commercial company and not, as the name suggests, an open source collaboration. Nevertheless, the video shows how current AI systems can offer a form of creativity. Watch the video before you scoff.
The agents eventually learn how to use tools and jump out of the game to score a win. As I found out years ago, the issue with such research is that the wanted applications are government security and the military. They could use it in positive applications, but humans are not like that. Guided weapons, NSA monitoring applications, and face recognition are immediately acceptable but don’t bother with medical diagnostics.
There is no breakthrough with OpenAI. Instead, the company seems to be using run-of-the-mill neural networks. Also, they have errors in their game programming, leaving openings for the AI to exploit. In a sense, we are watching how the AI system takes advantage of the limitations of the humans who set the game’s rules.
Their description is vague, as expected for a commercial venture. They appear to base the approach on a system called OpenAI Five. The main interest is that they throw masses of computing at the solution. Their neural system runs on 256 GPUs and 128,000 CPU cores with millions of tries. The system is brute force using unremarkable neural networks. Quantity can have a quality all its own.
The results are fascinating. This system may remind the reader of Stoffel, the honey badger, escaping from his compound. Stoffel plays similar games involving innovation and creativity. He cooperates with his girlfriend in escaping his compound. In Stoffel’s case, I doubt he uses anything like the OpenAI approach. Stoffel has a better “software” solution. He thinks about the problem and does not require millions of tries. Stoffel’s wetware is more efficient, and his internal modelling is genuinely creative. This creativity makes his solutions faster and more practical.
Stoffel uses better algorithms as he has to solve problems fast in the wild. Whatever methods his wetware is using, it’s efficient and effective.
We may contrast Stoffel’s fast single brain with the massively parallel system and long-time scales open to bacteria. Surprisingly microorganisms can be equally creative in solving problems. Like the brute force AI solution, bacteria may use simplified algorithms and compensate using quantity rather than quality.
Cancer
People misunderstand how cancer works. A tumour can develop over years. Damaged cells vary and divide relatively slowly, each division increasing the genetic and other variety. Many cells are not viable and die. Others find they are adapted to the conditions and divide further. Every division potentially generates a more able tumour cell.
The tumour is in a state of adaptation. It is a learning system. The cells are trying to work out a way of thriving. Now a small lump might contain a billion cells dividing for a decade. In other words, we have a brute-force learning system. The cells are figuring out how to escape the tumour’s environment and the host’s resistance.
Doctors describe cells that figure out the solution as malignant. But, like the agents in the OpenAI game, cancer cells figure out how to break out of the system. They escape the tumour, wander through the body, and set up new colonies.
Why have we made so little progress in treating cancer? Doctors have assumed cancer is just a dumb growth. The result is predictable. A treatment will fail unless it matches the variety in the cancer’s learning system. Even more critically, the therapy must adapt and model the cancer. Each cancer will be different, with different adaptations and vulnerabilities. The cells outsmart medicine rather like the game’s agents outmanoeuvre the OpenAI designers. Cancer’s difference is the cells are far more capable than the agents and equally underestimated.
Smart biology – it’s fun.