The Toddler’s 3 Rules for Understanding Artificial Intelligence and Machine Learning
“Having a two-year-old is like having a blender that you don’t have the top for” — Jerry Seinfeld
Any parent of a Toddler knows that they can be super smart — and in a very practical way. We moved to our new house a year and a half ago, and my son finally got his own space. We stacked a couple of low bookcases for him in his room, and put a cool train-themed lamp on top. He loved the lamp, but it wasn’t really kid-friendly — so it had to be put out of reach. Later that night I heard a crash in his room. My son, in the dark, had scaled 5+ ft of bookshelf and was playing with the lamp. So, what did I do? I pulled out some shelves to make it harder to climb. Next night, he had climbed that configuration too, in the dark. At that point, I just unstacked the bookshelves and accepted defeat. My toddler outsmarted me… So, I would say that Toddlers have a lot to say about Intelligence, and since they like robots, Artificial Intelligence seems like a small leap.
“I have not failed. I’ve just found 10,000 ways that won’t work” — Thomas Edison
Intelligence is Learning, and Learning is Experimentation
Let’s be honest, we “adults” are terrified of failure. Toddler’s are not. Toddler’s have a natural feel for learning and exploring their world, with little to no sense of impending danger or impending death. It never occurred to my son that climbing a giant (in his perspective) obstacle in the pitch dark was a bad idea. Frankly, the core job of the parents of Toddlers is keeping them from killing themselves in the process of their scientific experimentation.
When we are thinking about integrating Artificial Intelligence into the real world, we have to be ok with continuous experimentation, constant failure, and then inevitable, continuous progress. The wonderful Toddler process of small victories built on accommodated failures is at the core of machine learning in general. If we expect perfection as a starting, we cripple the whole enterprise. Of course, this is hard to sell to a world still fooled by the myth of leaps of innovation. The best evidence is success, so don’t be afraid to start small.
“Necessity is the mother of invention” — Plato
Learning needs to be practical, not just impressive
The AI news in the press is usually impressive and flashy (Like beating humans at Go or Reading), and… this doesn’t really help in the trenches. To really take off, AI must make an impact on the day-to-day minutiae of our lives — and that is definitely starting to happen (Think Digital Assistants like Alexa, self driving cars, or video games). More broadly, this is the problem with predicting the course of any new technology. Big ideas don’t always sell products and create adoption. The road to adoption is usually much more pedestrian than we like to admit. Just watch the Jetsons or Back to Future. They got it completely wrong because they assumed a straight line from where they were with the technology they had at the time. The problem is just it is nearly impossible to predict how the amazing technology we create will fare against the human race’s incredible stubbornness, as well as incredible imagination in using technology in ways the creators didn’t intend.
But back to the toddler. When the Toddler is learning, he applies himself to the problem right in front of his face. My son spends lot of energy figuring out how to best make mud pies, climb things he should not be climbing, and figuring out the tensile strength of his toys (I think he has a future in quality assurance). The Toddler sees no point in spending time on something that does not directly make his life more enjoyable, fun, or satisfying. I think one could argue that adults are the same — we just like to dress it up as “finding ourselves”. Bottom line, the successful AI applications will be those that actually solve real, practicable, and explainable issues for the average person.
“Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s?” — Alan Turing
It should be helpful, not a replacement
Have you ever asked a Toddler if they need help? “I do it” is the typical response (unless it involves bringing them snacks while watching TV, of course). The Toddler is open to assists and helpful hints, but don’t do it for him. It is really the same for us adults. People don’t typically like, and may even actively resist, technology that is threatening and seems like a direct replacement. While replacement is necessary and helpful sometimes (think manufacturing robots), it is usually better if the technology makes the correctly educated human more effective (think brake assistance and collision avoidance in cars). Particularly for highly skilled people, they want the Iron Man suit, not the pathological HAL 9000 (helpful, but in a very murdery sort of way).
The Toddler wants to feel in control and capable, more than anything else (honestly — that explains something like 90% of their behavior beyond eating and other bodily functions). Again, are the rest of us any different? We want technology to make us feel better and superior to those without the tech — original iPhone anybody? (seems like a lot of human behavior is about feeling superior to others… ok, this is getting too deep… pull back, pull back!!).
Bottom line, the Toddler is uninterested in your sophisticated notions and hoity toity ways. The Toddler, as the emperor of his tiny domain (at least in his mind), demands practical results and a clear path to useful outcomes like fun, laughter, and enjoyable destruction. And the deep truth is that we all have some Toddler inside — so it’s best to just accept it and move on…