Skip to main content

To AI or not to AI

Nowadays, everyone wants to paint a picture of AI. There are quite rational arguments being made, by thinkers great and small, about how AI will affect human existence - as it pertains to the replacement of our labor skills - and leave us with a deeply hollow middle.

Here are some concerns or rejections of concerns about AI that seem high on the list:

1. AI's promise of developing human style intelligence, or "how AI can never reach the level of human intelligence, because human intelligence is wetware and long-period-evolution ware". The concern, rightfully so, is whether AI will take away our general intelligence related jobs.

2. AI feeds off our data, and we need to take back control of our data and make sure the capitalists pay for it. The concern here is how money is driving AI in directions where the ethics of massive data collection and hoarding is simply not debated or discussed while leaving the decisions to "the inmates running the asylum".

3. Data is the new "oil", so we should demand that we get paid for our data and to keep our data under our control. The concern here is one of "data colonialism" where the data overlords take over the lives of the rest of us, by using us as raw materials to feed their colonial frenzy.

4. AI can take away our simple tasks, but it won't affect us where it comes to "real human intelligence". This concern discounts the real possibility that AI will do so, and needs a little more unpacking to understand why this stance will lead us to blindness of the eventual rise of real AI.

5. AI will make us digital hyper-connected sheep. This is a real concern, one that comes from a subtle misunderstanding of why AI is making us so [answer: it is our evolutionary heritage that is feeding AI in this direction].

These are some of the top concerns, but by no means exhaustive.

To be fair, these are valid issues. The solution is not clear to many, and we are still grappling with the sudden and sharp rise of AI's capabilities in the recent years (2013 and later).

The real concerns about AI are like history: they repeat themselves. Every new platform that democratized some aspect of human communication, trade or livelihood also consolidated power into the hands of a few. Technologies like AI are attractive to the very same forces in the government and illicit/dark economy, as they are to the capitalist/globalized data driven profit-seeking side of economy. This is the real issue that takes our control away from us, and one that should cause us to re-think our feelings towards AI.

I try to paint my answers as a different set of brush strokes below.

The fear of the new unknown

All technologies worried humans when they were newly discovered. What's different with AI is that its growth is fueled by extreme scale of deployment, speed over alternative solutions, and exponentially decreasing cost structure over time. 

AI is inevitable.

A lot of experts worry that our very self preservation is tied to an urgent and wide ranging appraisal of AI, with a view to regulate it before what they believe will be too late.

1. AI cannot be human intelligence

Matching human style of intelligence isn't necessary for other types of intelligence to displace human labor. Natural process of evolution resulted in a colony of cells with captive bacteria (mitochondrion), bacteria and virus (gut) to become individual units of existence by themselves - humans.

AI can be put together in similar ways - task specific AI mixed with meta AIs that write AIs, and so on, in a hierarchy of AI layers. This has already happened in some forms. 

AI, if truly developed with the best human capacities, will look very different from human intelligence. It will develop ways of thinking that might be equally other-worldly, and insightful in a way we simply cannot understand. That would be true success, although it has its perils.

2 & 3. Data is ours, take control

Data is oil. But data is just the oil. 

There are engines that use them: extremely complex black boxes designed by profitable companies, now spreading like wildfire to every business area. People love to hate them. But people also vote with their feet by using what's convenient and beneficial to their own pursuits.

Then there are the roads - highly connected networks with cores of super massive bandwidth. These networks simply grow by feeding on the traffic patterns of our lives.  They, in turn, feed our intrinsic need to connect, communicate and trade value. Physical labor powers most of our lives, but increasingly, information is driving the higher order bit in what we see as useful.

Sitting on top of this ecosystem is the human brain. A tight loop runs between human brain's capacity for infinite spans of attention-distraction, quick excitability when presented with frequent small rewards, and our cyclic/rhythmic fascination for the new.

There is digital gold in satisfying these higher order needs of humans with fast and cheap fulfillment through repetitive tasks.

4. AI cannot take away deep human tasks

It isn't a wonder that AI is taking over a lot of repetitive tasks traditionally done by humans.  Yes, most such tasks required a generalized intelligence to conceive, invent, experiment, sell and simplify. Now, machines and AI can do it cheaply and made available to everyone. The cellphone was a luxury just a few decades back. Now, it is indispensable to farmers and poor villages, as well as for the selfie loving teen, or the researcher, or the e-delivery service provider.

5. The digital hyper-connected sheep are our biggest issue.

The real challenge from data+AI is that the massive collection of data attracts very different scale of consumers to the party in a way never possible before privatized data gathering on global scale.

For large actors like nation states, cartels and enforcement departments, the possibility of spending very little of their money to get at the data of their constituents and competitors are huge. They have the right incentives - large payoffs and self-preservation. They would be losing out to their own competitors, if they stop this data fire-hose.

Soon, all the data need not come from humans, since AI is now able to generate data for training and imaginative extrapolations of human inputs. Forget the IOT, there's an IOI (internet of intelligence) that will be generating its own universe of simulations that will have the potential to help (or mar) us even more.

Calm before the sunrise.

The politics of AI will need the best and brightest - students, teachers, engineers, political scientists, writers, critiques, legal scholars, economists, philosophers, scientists, workers and leaders - to understand, extend, and put it to beneficial use.

We won't be hyper-connected sheep. There will always be the hyper-connected addicts and the hyper-connected hyper-successful. But AI can likely impact the not-so-hyper-connected masses in positive ways if multi-disciplinary effort can understand how to make the best use of it.

AI is in the early seconds past midnight. Sunrise is a long time away, and whether we use it or burn from it is entirely up to us.


Popular posts from this blog

Why PI is not 4, math is great, and other mysteries.

The other day, I found myself with an interesting problem of approximating a circle with the enclosing square which seems to prove pi = 4.

The paradox was forwarded by a most interesting puzzle collector, Surajit Basu, a friend and life long inspiration. See Sonata for Unaccompanied Tortoise for why!

Here is the offending paradox:

This is an example of how counterintuitive questions can be answered with a little calculus.

The key is to realize that no matter how closely we approximate the circle, the orthogonal lines of the approximation formed by inverting the square corners will never actually be tangential to the circle.

Note carefully that as you get closer to 90 degrees, the horizontal line is much longer than the vertical. Same goes with the approximation at 0 and 180 - the vertical line is much larger than the horizontal component.

If we take a quadrant of the circle - let's say the top left quadrant, moving counter clockwise from top to left -  we can imagine that each inf…

Ambition vs. Fear.

Most important things in life don't come to us. Nor do we get them by seeking/wanting them. It comes from letting go of the unimportant stuff.

The hardest part is letting go of the tendency to take the world as is. This is a habit of our past successes.

But success is not a destination, it is a STOP sign. You stop, wait, and move on. Too often, we are paralyzed by success into the fear of the new. We stall on the road to a new life. We need to break our inertia and move.

Our thoughts and thought habits are hard to break. But that is where we have to spend the most energy. Thoughts are always competing strands  - of worries of the past and anxieties for the future. For some of us, they are cleanly separated into rivers that nurture every place they travel. For most, they are like the torrents and trickles -- competing, rushing somewhere, stopping completely elsewhere, always mixing, morphing, competing, winning, losing.

Our thoughts are the potential difference between the two pole…

Architecture, Engineering, Operations - iteration 1

The world has infinitely more stuff to be "done" nowadays. At least in the sense of building/running an institution that uses technology, there are many roles that are involved in making things work. The world of IT and technology in general makes the speed and variety possible. We now have a platform of IT that is globally scale-able if we can put some new thinking to the old problems of "getting things done".

There are great organizations that do this well, and they use "modern" IT principles to achieve this.

Fundamental to engineering a modern IT (or infrastructure organization) are the three roles of Architecture, Engineering and Operations. Some would say Architecture is encoded Engineering-history, but for now, we will keep them separate.

The popular definitions for these roles are about "output" delivered or the "domain" of discourse. The personality drives that determine the actual performance are not discussed, as far as I ca…