I’ve been unusually interested in artificial intelligence lately. To where I’m freaking out some of the people I’m regularly around. A lot of my interest stems from wondering what are the limits of AI? Whose jobs will it replace? Is there any job it one day won’t be able to take?
While watching the Super Bowl, where drinking beer and eating Crustables peanut butter and jelly sandwiches was more entertaining than watching the game, I started reading about it again. My girlfriend, who was incredulous I’d be reading about artificial intelligence during the Super Bowl, starts talking to her niece instead of me. Her 8 year old niece goes, “What is he reading about?”
Girlfriend, “You should ask him.”
Niece, “What are you reading about?”
Me, “Robots and artificial intelligence.”
Niece, [Stares for a while] “Why are you reading about that?”
Me, “Because, I need to be prepared if they take over. Like, what will we do? What happens if robots are smarter than us? Where they can do all our jobs? Can they ever do that? Can it go further than that? Is Skynet plausible? I don’t know. What do you think?”
Niece [Stares again…Walks away]
The next day my girlfriend and I are at Panera. She’s ordering and bantering with the cashier. I’m watching the girl take our order, poking various buttons on the screen in front of her, she takes our card, swipes it, gives us our number and we’re off.
Me, “Now, why is she standing there? We saw what she did. She doesn’t technically need to be there. I can poke all the buttons she poked. Or when we go to the movies. Why does no one use the kiosk but us? 95% of the people go to the cashier.
Girlfriend, “Because people want human interaction.”
To my girlfriend, who manages a bunch of stores regularly dealing with face to face customers, this is obvious. Considering my profession, I suppose it should be obvious to me too. But, considering I sometimes enjoy dealing with dogs more than strangers, I often opt for the kiosk.
Turns out, in learning about AI I discovered something called Moravec’s Paradox, which is considered one of, if not the, biggest revelation of artificial intelligence so far.
“Moravec’s paradox is the discovery by artificial intelligence and robotics researchers that, contrary to traditional assumptions, high level reasoning requires very little computation, but low-level sensorimotor skills require enormous computational resources.”
Sometimes your random interests can lead to a great insight. I think this was one for me.
Let’s break this into two domains:
1) High level reasoning
- Logic exercises
- Mathematics
- Engineering
- Playing chess, checkers, cards, etc.
- Analyzing stocks
2) Sensorimotor skills
- Recognizing someone’s face
- Walking, especially in a coordinated manner. (Not bumping into things.)
- Writing with a pen
- Knowing where your body is in space
- Shooting a basketball
- Taking an order at Panera while talking about how awful the Super Bowl was the night before
More broadly, high level reasoning tends to be things we really need to think about, where sensorimotor skills are things we don’t pay much attention to. There’s of course variance here. After years of practice you don’t have to heavily think about how to add numbers anymore (unless they’re big). Whereas if you’ve never shot a basketball, you’re going to be thinking a fair amount.
The paradox is what we humans think requires a ton of brain / computational power, such as high level reasoning, actually requires little computational resources. It’s easy for a computer to not only do your multiplication table, it’s easy for it to multiply any number you can think of. 218 * 329 = “Hard work” for humans, but easy for a computer.
On the other end, what we consider to be rudimentary, gardening for instance, is extremely hard, if not impossible, to get a computer to do. While my girlfriend and I could technically go to Panera and order without talking to a person, Panera can’t, through computers, generate the kind of human interaction most people enjoy when ordering. If a business could cut costs and not hurt revenue by taking people away from the cash register, I think they would have done it by now. My hunch is they already intuitively know that’s not a good move, or they found out the hard way (lost money). People matter.
It’s like when you go to a bar. We have the technology for a machine to fulfill your order. What we don’t have the technology for is recognizing you’re a regular, listening to your belligerent self ramble about life, understand what you’re saying, and converse back and forth with you. All done while actually watching the baseball game.
In college, during my math majoring years, I was exposed to a software program called Mathematica. Upon using this I was completely blown away and completely incensed at the same time. This program could do ANY math problem you put in it. Screw addition or subtraction, this thing was doing triple integrals. Consider how many people think algebra is hard, which is often taught to 8th graders. We have computers which can do Calculus 3 problems -reserved for upper level college students- as fast as you can type them in. The type of math we consider only for people with IQs in the, I don’t know, 95th percentile?
Yet, we give no credit to the airline customer service representative, who can talk down the asshole blaming them for it snowing. The computer can easily do what we think only 1-5% of people can do; the computer can’t do what we think anyone can do.
After learning about Mathematica I left the class thinking, “You’re telling me I’ve spent 15 years getting good at this crap, meanwhile a computer can do everything I’ve learned? Why the hell are we learning this then?” You know what’s even worse? This thing is available as a free website now: www.WolframAlpha.com
Hence, this quote from Steven Pinker:
“The main lesson of thirty-five years of AI research is that the hard problems are easy and the easy problems are hard. The mental abilities of a four-year-old that we take for granted – recognizing a face, lifting a pencil, walking across a room, answering a question – in fact solve some of the hardest engineering problems ever conceived… As the new generation of intelligent devices appears, it will be the stock analysts and petrochemical engineers and parole board members who are in danger of being replaced by machines. The gardeners, receptionists, and cooks are secure in their jobs for decades to come.”
I bring up my math days to illustrate we need to rethink what we consider is work for our brains. I spent two decades listening to people bitch and moan how hard math classes were. But Mathematica could do any problem our brains could do, do them all faster, and never make a mistake. You know what Mathematica couldn’t do? Walk up a flight of stairs.
Just because what you’re consciously doing seems hard, doesn’t mean what you’re unconsciously doing is not hard. Difficult is relative. We often perfect skills until they’re a habit; what happens when a skill becomes a habit? You don’t have to think about it as much, if at all.
Moravec has used evolution to understand his paradox. Mathematics, reasoning, engineering, abstract thought, are all new to humans. Maybe 100,000 years old. On the other hand, sensorimotor skills have been around -at a minimum- a billion years.
Think of it like any skill you’ve acquired. When you first learn it, it can take a while to grasp. But, once you’ve had the skill for a while, it doesn’t take much conscious effort to complete it. Like riding a bike. That doesn’t mean your body isn’t doing a ton of unconscious work though. After the learning hump is over, you might not need to focus on how to push the bike pedals, but pushing them requires a lot more resources than doing a crossword puzzle in your leisure time.
We’ve been using our sensorimotor skills for a long time, so we don’t think about them much. We haven’t been using our reasoning skills that long. So, to humans, it seems like a lot of work. It’s like we’re in the tricycle phase of learning to use our reasoning skills.
In other words, something feeling difficult in a relative sense doesn’t guarantee it being difficult in an absolute sense. No matter how much you focus on a math problem, it’s unlikely to ever work your body (brain!) as much as walking. But, we don’t think about it this way, we only think about how we feel. Unless they’re decrepit, nobody says “Walking is hard work.” Unless they’re insane, nobody says “Math is easy.”
So, what do we do when someone is showing signs of cognitive issues or cognitive decline? We make them do versions of math, logic and reasoning. Like “brain games.” We think this is what challenges their brain most.
Yet, what requires way more computational resources? Doing something like moving their body around. Which is why we find exercise trumps mental activity in warding off dementia. In fact, exercise can make the brain bigger. Brain training, at best, appears to prevent how fast the brain gets smaller. It does not improve the size of the brain.
I haven’t trained a ton of people with dementia, but I’ve worked with a decent amount. When I think about this stuff two people in particular come to mind. One who was a former member of the CIA, and another who was a doctor for 30 years. When it comes to the brain, we all hear “If you don’t use it, you lose it.” Based on our definition of using the brain, it’s challenging to argue these people weren’t using their brains enough. It’s tough to be a doctor without regularly engaging your mental resources. But, it’s conceivable they were not using their body enough. It’s easy to be a doctor and not be active.
Again, the brain needs a hell of a lot more oxygen when you’re walking for 30 minutes than it does when you’re doing Lumosity.
Even other things we associate with improving brain function, like being social, it’s not clear that helps either. The first study I linked found:
“On the other hand, our study showed no real benefit to participating in mentally and socially stimulating activities on brain size, as seen on MRI scans, over the three-year time frame.”
Another study did find being social helped with cognitive decline:
“He [the researcher] asked them about going to restaurants and sporting events, playing bingo, doing volunteer work and other activities.”
Notice what the above instances consist of. They all require more activity. It’s not only, “Those who sat around talking with people.” To socialize these people had to get their ass up, go somewhere, move around wherever they were, and socialize within that. Is it the talking to people that helps, or is it the movement?
This is a theory I have with coffee, which has been found to help with practically any ailment, including cognitive decline. What you often find is, up to a point, the more coffee people drink, the better off they are. This is exactly what you find with physical activity. More physical activity, up to a point, makes people better off. (There’s always a precipice of negative returns. Though, 99.9% of people never reach this.)
We all know what caffeine and coffee does to us. One of the most common traits is it makes you move more. It might only be twitching your leg, but that’s still moving more.
Rather than some complicated chemical pathway benefit from caffeine, to me, you can reorganize coffee drinkers as:
- Drink none; move X amount.
- Drink a little; move X amount + a little more.
- Drink a good amount; move X amount + a little more + even more.
Alzheimer’s is more prevalent amongst minorities than it is caucasians (here). A known difference between these groups is minorities exercise less than whites. This also makes sense as we’re finding heart health correlates well to brain health. (Minorities are at greater risk of heart attacks and other vascular issues.)
Alzheimer’s is more prevalent the older we get. What happens as we age? We move less. Those over 65 are the least likely to exercise (here). I realize there are a lot of other factors to consider, but there’s even research to suggest, with a linear relationship, the more older people move, the less likely they are to get the disease.
“Over a mean of 3.5 years of follow-up, 71 participants developed Alzheimer’s disease The research found that people in the bottom 10 percent of daily physical activity were more than twice as likely (2.3 times) to develop Alzheimer’s disease as people in the top 10 percent of daily activity.
The study also showed that those individuals in the bottom 10 percent of intensity of physical activity were almost three times (2.8 times) as likely to develop Alzheimer’s disease as people in the top percent of the intensity of physical activity.”
The cool thing about the above study is it didn’t only measure exercise, it measured movement. Had to get up and cook? That counts, and that helps. This is better than a lot of other research, as many studies look only at a subject’s exercise, which is often only walking.
It’s worth noting, research where walking was the primary or only exercise still outperforms “brain training.” Yes, WALKING is outperforming what neuroscientists are coming up with. What is close to the most basic thing we can do, walk, outperforms what our societal geniuses can formulate.
What if these subjects diversified the motor skills they focused on, doing things involving switching from right to left, ones with increasing levels of coordination, where they are also in a social environment (e.g. with others or have a trainer), exercise which is progressively loaded…I’m betting this would outperform walking.
Not only do the above heavily work on sensorimotor skills, they involve thinking and learning. Doing a new exercise becomes a new skill acquisition. One in which they could be interacting with someone (social), have to execute things in a specific manner (coordination), are trying to remember the exercise (memory), have to understand where their body is in space (kinesthetic), are progressively challenged (greater neural output), the benefits are endless.
Hell, walking in a foreign area involves much of the above. When you really think about this, it’s pretty clear how going for a stroll with a friend involves more mental resources than picking squares on your computer screen. The intentions at places like Lumosity are great, but they aren’t getting the job done. Besides, going for a walk with someone is better than spending even more time staring at a computer screen.
–
Posted on February 6, 2014