On Monday, I posted an article titled The AI Misinformation Epidemic. The article introduces a series of posts that will critically examine the various sources of misinformation underlying this AI hype cycle.
The post came about for the following reason: While I had contemplated the idea for weeks, I couldn’t choose which among the many factors to focus on and which to exclude. My solution was to break down the issue into several narrower posts. The AI Machine Learning Epidemic introduced the problem, sketched an outline for the series, and articulated some preliminary philosophical arguments.
To my surprise, it stirred up a frothy reaction. In a span of three days, the site received over 36,000 readers. To date, the article received 68 comments on the original post, 274 comments on hacker news, and 140 comments on machine learning subreddit.
To ensure that my post contributes as little novel misinformation as possible, I’d like to briefly address the response to the article and some common misconceptions shared by many comments.
Welcome to the Club
Regarding the role played by the popular media, one common (and reasonable) response was – hey, this happens everywhere. When you’re an expert, then you realize the press doesn’t know much.
In an unusually eloquent comment, a man going only by “Jim” wrote:
This is a problem that impacts every area of expertise. Journalists are, for the most part, generalists rather than specialized correspondents on a particular area. -Jim
I buy this argument but think it requires qualification. As a jazz musician, I often lamented the failure of the press to cover the substance of the music. The only jazz writing that hit the spot for me was that by pianist Ethan Iverson (see his Do the M@th blog) who happens to be an uncommonly gifted writer in addition to being a prolific pianist (as featured in the Bad Plus).
So, in short, I don’t feel sorry for us machine learning researchers who don’t always get accurate coverage. This is a common problem. And in fact, in the short run, many of us benefit. Salaries and startup valuations are at stunning levels and the incredible demand from students (much of it fueled by media attention) for master’s and PhD degrees in machine learning in turn has created demand for machine learning professors and lecturers. However…
The Potential for Harm
Putting aside sympathy for the machine learning community, there’s another more important concern. Unlike string theory, where a cartoonish depiction of research doesn’t change anyone’s life, machine learning actually is affecting people’s lives. We interact with it every day, and the problems are real, e.g. (i) discrimination in algorithmic decision making (ii) technical unemployment (iii) ceding control to mindless recommender systems that optimize only clicks (think Facebook’s fake news problem) (iv) autonomous weapons already under development raise serious concerns about the ethics of warfare. So machine learning will have both immediate and long-term societal impacts, and it actually is important, for democracy to function, that the public be better informed.
In my eyes, the present situation is reminiscent of the news’ failure to adequately cover the derivatives market prior to the financial collapse. How can we hold politicians accountable for reasonably regulating the financial markets if Michael Lewis is the only pundit who understands how the derivatives markets work and it takes him over a year to write the book that explains it for the masses? [TL/DR: cloying prose but clear explanations]
Religion and the SingularitY
Some readers were peeved that I referred to Kurzweil’s Singularity prophesying as religion. It’s true that in scientific communities, calling something religious can function as a Bogeyman argument. But I feel comfortable using the term here. Kurzweil has a conclusion from the outset. Man and technology will merge – the technology will explode doubly-exponentially (whatever that means), this will result in him living forever, and it will happen at a date that only he can deduce by some opaque means. This has undeniable hallmarks of religiosity.
- A holy figure / cult leader – Ray Kurzweil
- A nebulous prophesy – The Singularity, in its evangelized form, is defined such that a true believer can always claim that it has already happened and always claim that it hasn’t happened yet, and yet (of course) it is prophesied to happen at a precise date: recently revised to 2045, as gleefully reported by the Telegraph
- Unshakable axioms – Scientists fit beliefs to evidence. But Kurzweil’s Singularitarians cherry pick evidence that fits beliefs. See the strange conversation regarding doubly-exponential growth of “technology” in the original post’s comments. What precisely does technology mean here? This can shift over time (or even within a single graph), as long as something appears to fit this unshakable portion of the sacred text.
Some readers also pointed out that ‘singularity’ means many things as defined by many authors. Because the post addressed the misinformation epidemic and Kurzweil’s variety has a near monopoly on the public consciousness, I was speaking specifically as regards this flavor. Perhaps in a future post I could review the ideas of other, more critical thinkers on the topic more thoroughly.
Also, some readers took my dismissal of sensationalism and quasi-religiosity as a dismissals of either (i) AI safety, (ii) futurism, (iii) the dangers of unregulated development and deployment of machine learning.
To be clear, I make no such dismissal. Broadly defined, I believe that developing a rigorous study of AI safety is of fundamental importance. Already we live in a world with autonomous drones, autonomous automobiles, and one in which the military is developing autonomous weapons. Unchecked applications of machine learning have the potential to cause physical, financial, and social harm (see above).
Per Merriam Webster, a futurist is defined as:
one who studies and predicts the future especially on the basis of current trends.
In this sense, I would consider myself to be a futurist, although perhaps more focused on the near to intermediate term than the typical futurist. I have no problem with futurism, as long as one reasons clearly and logically.
Just an Outline
Some readers pointed out that the post was just an outline of works to come. Yes, this is true.
Diversity and Inequality in AI/ML
Finally, while witnessing the traffic bubble to Approximately Correct was admittedly gratifying, a disturbing trend turned up in the traffic stats. For the better part of a day or two this article was a top trending piece of machine learning news. In that time it received nearly 16,000 views from the US, 5,000 of which came from California. Traffic numbers from western European nations ranged from the hundreds to thousands each. However, the total view count from the entire African content was less than 300. Excluding South Africa, that total falls to less than 100. As we develop technology with the potential to displace entire industries, including the agriculture, mining and transportation industries that many developing nations depend on, we should ask who owns / should own this technology? And what will happen if we decimate the employment markets in countries that reap none of the benefits of such ownership?