A few thoughts on AI
Writing about Artificial Intelligence is tricky, because it is so trendy to talk about it these days. Yet the victories of AlphaGo against Lee Se-Dol have fascinated me and prompted this writing.
The AI victory at Go is fascinating because DeepMind seems to have developed its own intuition
The AlphaGo victories appear to be a breakthrough in this history of Artificial Intelligence, not least because in order to win, the program had to rely on some sort of intuition. Go was known to be too complex of a problem to be approached in the “brute force” fashion. This traditionally involves modelling all the possible outcomes of a finite game and choosing the path that leads towards the desired outcome. As many have written, AlphaGo does not rely on such approach. Instead, it learns to recognize patterns on the board to identify subsets of all potential plays, and analyze in greater depth those subsets that appear to be most promising. It also trains itself by playing against itself, millions of times, leading it to play surprising moves, even to some the game’s best champions. Yet, akin to the programs being developed to drive, cook, or recognize images, the fundamental principle remains the same: developing decision-making capabilities based on pattern recognition.
All pattern recognition tasks may one day be performed by programs
Many of the human tasks we do today are indeed pattern recognition tasks. These can be fairly “simple”, such as hand-writing recognition, matching a pattern to a codified set of letters, then themselves matched to concepts (words and their definitions), or can be more complex, such as driving a car or diagnosing the condition and optimal treatment for a medical patient, or playing Go. In all these situations, the technology is there to make programs as good, if not better, than humans. There are remaining obstacles in the three stages of the process (namely receiving information, processing information and sending a response of the right format, such grabbing an object, running, whistling, any sort of actions humans take in response to an information signal and its processing), but these are likely to be overcome over time, as hardware and software become cheaper and better. AlphaGo is yet another sign of this.
Over time, this is likely to result in the automation of many tasks humans do today. Some see see this as a negative consequence. However, I do not think it is. There is the traditional economic argument that having these tasks performed by machines will make them cheaper and therefore enable trade of more services and products. I also think that a world where humans do not need to perform the repetitive tasks they are currently doing is a world where humans will be able to focus on other, better, more interesting things.
Humans will have the opportunity to focus all their time on non-pattern recognition tasks, such as imagining, creating, etc
Creating, inventing things, reading, writing, imagining, telling stories, these are just a few of the things I would do if I had more time in my day — all the time that I am spending on recognizing patterns that I would happily trade for these more creative activities.
This does not make me a naive optimist. The transition from an economy based on “pattern recognition” jobs to one based on heavily imaginative jobs is likely to come with pains and challenges. History also suggests no intrinsic reason to believe a heavily imaginative economy would by itself be good. Many people will attempt to create or imagine things, or tell stories that benefit only them, at the expense of others.
However, regardless of the good or evil uses different people will make of these changes, this suggests what’s coming is not so much the end of humanity, but yet another exciting stage of it.
A few ideas to prepare for the world to come
It also suggests areas to invest in for humans who want to anticipate the change. This deserves another writing session, but I will just put down a few that come to mind right now:
- The ability to really think about new things will become more important for humans — as opposed to the ability of performing tasks (such as driving, producing a diagnosis, playing a game) again and again to be good at it. This means this ability should be trained both by maximizing the exposure to imaginative activities, and minimizing the exposure to activities where the mind is kept captive (such as endlessly scrolling through a Facebook or Instagram feed without really thinking nor resting the brain)
- An understanding of programs will become more important, not so much being able to code, this is probably going to be automatized as well one day, but the ability to understand what programs can do to turn a vision into action
- The ability to engage with other humans will become more important. As we do less and less of the repetitive, solitary tasks, and the value of humans shift to imagining and creating, being able to share, work in groups, exchange ideas will become more important. Great ideas rarely were born from a man alone. At the same time, the ability to engage in productive, content-rich conversations as opposed to simple gossip will be important too (although gossip itself may well be what makes us humans, as suggested by Y. N. Harari’s in Sapiens: A Brief History of Humankind, and I am not suggesting we should end gossip altogether)
Schools will have a role to play in teaching these abilities. In many countries, this is a pretty big stretch from the current school model, focused on the accumulation of mostly theoretical knowledge rather than creativity, collaboration, and being technologically savvy. But luckily for those already out of schools, or those who don’t want to wait for schools to change (this may take a while), it has never been easier to read about pretty much anything and share your creative ideas with other people.