Tribbles, Turing & TPU2

Before reading, note that the topic of solipsism is discussed somewhat in earnest herein. A bit of rambling should therefore be expected.

I suffer from aspirational polymathism. It is a disease that has plagued me since my earliest memories. As a child I was not content to study just one or two subjects. I craved constant stimulation from a wide array of new subjects, new ideas, new challenges, and asipred to master them all. Perhaps had we had more of the prescriptive remedies so popular today to address such childish thinking, I might have been diagnosed as hyperactive or ADD, been given my control meds such as to dissuade such childish ambitions. Then perhaps I would have had less interest in such a wide and crippling array of fields of study. And I might have learned to appreciate golf. But those were less sophisticated times. Alas, I accept my handicap and do not allow it to disturb my otherwise placid nature. In fact, I am fortunate to have found an avocation that sometimes even lauds this general scholar disease.

One of the attributes of aspirational polymathism is a constant need for the stimulation from new and deeper subjects to study. And it can’t be just one at a time. At any given time, I don’t have one book I read on the Kindle app on my iPad, but generally three or four open and active on my queue, processing in parallel. I read, switch off to another, read more, and so on. Plus I have NetFlix or Amazon Prime or HBO running in the background, sometimes with simultaneous video streams from one course or another, sometimes working on code or email or PowerPoint, all at the same time. It can be maddening, but also rewarding. Sometimes I fool myself into believing that I actually get more accomplished by giving full vent to my disease. I fool myself into thinking I learn more and faster when I indulge my manic mode.
I confess that I do also enjoy an over-the-counter nootropic stack of my own concoction, sort of a Limitless fantasy I have, pursued for the past couple of years, and frankly it does seem to help my focus considerably. But aspirational polymathism does come with its own set of drawbacks. One of which is a tendency to jump very quickly to consider myriad down stream probable outcomes, with bifurcating branches of complex scenarios, most of which are way out of main stream thinking, as it were. My initial take on Google’s first TPU announcement, for example, was one of those moments. The more recent emergence of TPU2 and all it implies is another. But first a little more background.

So I have been reading SuperIntelligence and at the same time reading a text on Quantum Mechanics, and at the same time digesting Alan Turing’s 1950 paper for an AI course I was auditing online. Although I had read about Turing’s paper many times and read reviews and critiques, I had never actually read the original.

Alan Turing

Turing’s paper is probably the cornerstone of all Artificial Ingelligence R&D today, if not also the backbone of Theoretical Computer Science. I remember something Scott Aaronson wrote about Turing’s paper — how 70% of AI today can be traced back to that work. But it’s not the Turing Machine, nor the Church-Turing Thesis, nor the Imitation Game itself which struck me as especially germane from Turing’s paper. All those innovations I have read about, considered, appreciated, and more-or-less learned over the past few decades. No, it wasn’t all that from Turning’s seminal paper. It was something very different. It was statements he made about solipsism. The fact that he used the term 3 times in the paper is probably not all the surprising given the nature of the question: Can Machines Think? But When presenting his counter-arguments to those who, at the time, would out-of-hand dismiss the very question, Turing outlines a series of arguments to refute the detractors. He categorized objections based on perspective, for example, theological objections, or mathematical objections. One in particular, from the perspective of consciouslness, caught my fancy.

Can machines think? First we need to stipulate what we mean by ‘think,’ which is not as straight-forward as one might….think. So what does that mean? In denying the validity of the Imitation Game (i.e.: the Turing Test) the objector in question, Sir Geoffrey Jefferson, a British neurologist and pioneering neurosurgeon, expressed the view that writing a sonnet or composing a concerto, based on thoughts and emotions felt, and was the basis for judgement; and a machine was incapable of such feeling, and therefore could not think. Per Turing:

“According to the most extreme form of this view the only way by which one could be sure that a machine thinks is to be the machine and to feel oneself thinking. One could then describe these feelings to the world, but of course no one would be justified in taking any notice. Likewise according to this view the only way to know that a man thinks is to be that particular man. It is in fact the solipsist point of view. It may be the most logical view to hold but it makes communication of ideas difficult.” (emphasis mine)

With humor and brilliance Turing refuted Jefferson’s argument; taken to the logical conclusion, Jefferson’s view necessarily leads to hard core solipsism. It may be the most logical view indeed. But, as Turing observed, the motivation to communicate just about anything is a bit hampered when sentience is not presumed to exist in the ‘other.’ Therefore, let us stipulate that other human beings, despite frequent evidence to the contrary, do actually think. Further, let us accept the evidence of thinking to be communication itself. I believe Wittgenstein would agree. Without communication in some manner, evidence cannot be gathered.

I believe the Philosopher Wittgenstein’s influence on Turing is in evidence here. Turing did attend lectures by Ludwig Wittgenstein at Cabridge in 1939. Per notes from students attending those lectures (another book I read in tandem), Turing and Wittgenstein enjoyed many robust exchanges. Given the Philosopher’s views on solipsism and the limits of understanding, it follows that Turing was influenced in some ways by the assertion that the limits of language means the limits of my world.
Ludwig Wittgenstein
If thinking is manifested by the process of communication, indicating will and purpose, then we might agree that human life at all levels implies some modicum of thought. We ought not conclude otherwise. So to Turing’s question: can machines think? The question necessarily follows: can machines communicate? The two, per Wittgenstein, are linked.
So now let’s consider TPU2 and all it implies. In the ASIC v. GPU battle, even with awesome new GPU options, ASIC, at least for TensorFlow, will always win. Google’s TPU is TensorFlow burned into a chip. That’s great. Cool stuff. So where can I buy one? Where might a get a rack of TPUs to power my next AI startup? Answer: you can’t. You have to rent them from the Google Cloud. Maybe that’s not a big deal. But maybe it is.

As the market for deep learning grows, if it grows at levels predicted, the differentiation provided by the TPU will likely suffice to give Google an edge akin to that which it already enjoys in search. No need ot create and sell hardware — just rent out intelligence and take for Google mass user processing and data to sweeten the deal. I’m not sure when the “Don’t be evil” line gets crossed, but something tells me we may be getting close.

By the same token, I’m a big fan of TensorFlow. It’s awesome. This brings up the Tribble metaphor.
The Trouble with Tribbles
You remember Tribbles from StarTrek of course. Cool, happiness-inducing, soft and sweet little pets. Everybody wanted one. One arrived on the Enterprise and the entire crew was enchanted by the sweet Tribble. In relatively short order, the little thing made a baby Tribble. Now more of the crew could pet one and love on one. Then the two Tribbles gave rise to four….and so on. You get the moral of the story, yes? Exponential growth, even of really cool stuff, might have extremely deleterious unintended consequences. Our TPU enchantment may give rise to some really sweet and productivity-increasing applications — stuff we can’t yet even imagine in this early chapter of the innovation-galore Network Age. But beware the Tribbles. They may be hiding in our closets or under our beds.

Closing this entry, it was Wittgenstein who said, “Nothing is so difficult as not deceiving oneself.” The same is true of the aggregate of ourselves. I am quite sure I deceive myself with my aspirational polymathism, and I am not much more than borderline trainable and merely opinionated. And I am also quite sure, besides me, other people actually think. If there’s hope for humanity, it’s in software, solipsists notwithstanding. Can machines think? Perhaps we will know soon enough. But beware the Tribbles. They too come with progress.

Posted in Big Data

Leave a Reply

Your email address will not be published. Required fields are marked *