George Rebane
[This is the addended transcript of my regular KVMR commentary broadcast on 13 February 2019.]
The entire field of artificial or machine intelligence (AI) seems to be more misunderstood by the day, and not only on Main Street, but also in the highest levels of government. President Trump recently signed an executive order to launch the American AI Initiative. The purpose of this whatever it is, is to “focus federal government resources to develop AI” in order to “increase our prosperity, enhance our national and economic security, and improve quality of life for the American people.”
Now AI has been in steady development and use across the land for at least 30 years. Launching AI is not the same as launching a rocket to the moon when no one had done it before. The launch of AI decades ago was the same as the launch of, say, relational databases, spreadsheets, the microcomputer (we don’t call them that anymore), browsers, cell phones, and, of course the World Wide Web. All of these things started off bit by piece, then gained momentum in the market place as people saw new uses for them, and finally took off and became established in our businesses and private lives. AI has been no different, save in new areas such as the much ballyhooed, super-human game players, self-driving cars, and automation from factory floors to offices all over the world.
The main difference about AI is that it has the ability to become the greatest job killer, money maker, and inequality promoter in human history. Why? Because no one sees any limits to the intelligence of these machines combined with the agility, speed, and power of their robotic versions. It’s not clear that the federal government even knows what this AI Initiative is actually supposed to do. The PR copy states that “federal agencies will increase access to their resources to drive AI research by identifying high-priority federal data and models, improving public access to and the quality of federal AI data, and allocating high-performance and cloud computing resources to AI-related applications and R&D.” But industry, academe, and entrepreneurs are already doing all that at a tremendous pace and on their own, thank you very much.
The feds claim that with their “proper leadership, AI can empower American workers by liberating them from mundane tasks.” What this initiative doesn’t seem to understand is that such workers will also be liberated from their jobs, simply because they cannot be trained to do the more mentally demanding work. There is a reason why most of them now have mundane jobs, along with the over 7M job openings today that require skills unavailable and, sadly, unachievable by most of the unemployed. As more workers are replaced and become unemployable, systemic unemployment will grow – meaning there will be more unskilled humans than human jobs, at least the kind that pay enough to maintain an acceptable quality of life.
This AI initiative seems not to have been well thought out, and promises to add more bureaucracy and burdensome regulations to hinder new and innovative applications of AI. The main author of this initiative is the Deputy Assistant to the President for Technology Policy, Michael Kratsios, who is not a scientist or engineer, and seems to have a very incomplete understanding of AI, its development, and how such technologies arise and merge into markets. His training is in political science with a certificate in Hellenic studies. His recent article in Wired magazine titled ‘Why the US needs a strategy for AI’ clearly shows that he is navigating in waters new and strange to him. Of course, such forays into the unknown by government bureaucrats is nothing new. Mr Kratsios really thinks it will be the federal government that will “craft guidance for AI technologies that will promote innovation while respecting privacy, civil liberties, and American values.” And it gets better – the authors of the American AI Initiative actually believe in something called “the race for AI”, one that “we will win” under the shrewd guidance of the feds. And this victory will be ours “without compromising our American values.” Oh my.
Anybody who has participated in the development of leading-edge technologies, or at least is familiar with the history of such developments, knows that government regulators will not be able to control and/or guide innovation in the field of AI – that cat is long out of the bag. From here the government will do well just to keep up and understand what AI is being invented and applied, so that they can fund and acquire the proper technologies in a timely manner to fulfill their national security mandate.
My name is Rebane, and I also expand on this and related themes on Rebane’s Ruminations where the addended transcript of this commentary is posted with relevant links, and where such issues are debated extensively. However, my views are not necessarily shared by KVMR. Thank you for listening.
[Addendum] Some more links for the American AI Initiative. Forbes’ critique of the initiative come in the ‘what could possibly go wrong’ and ‘where’s the funding’ versions. Vox continues their broad-brush, general purpose TDS coverage, and also notes Kratsios’ lack of science training (here). Mr Kratsios background is expanded here.
To mention some specific job categories that will be toast in the near future, consider professional transport drivers on our streets and highways. Today, there are over 3M long-haul truck drivers employed to keep the wheels of commerce turning. And for some more white-collar oriented redundancies, consider the several hundred thousand portfolio managers along with more thousands of stock bonus program administrators who will be replaced by much smarter and cheaper software. The takeaway here is that a broad swath of middle skill-level jobs will fall to intelligent systems. No one has yet come up with a viable Plan B for these folks.
Ethics and political correctness in AI implementation is something that a lot of the Left is getting worked up about – specifically, we mention the entire notion of ‘AI bias’ (see the Forbes piece for an expansion). Since humans are NOT created equal along any of our defining attributes, deep learning AI programs will find such discriminating patterns in the mounds of ‘biased data’ that will be used to train them into final algorithmic forms for fielded systems in commerce and government. And bureaucrats (in the AI initiative?) will demand to have additional work to do adjudicating such systems for their bias before they are allowed to be deployed and used in the worldwide fields of administration. Detecting bias is hard, but easier than proving that any given algorithm (that’s what computer programs are) is NOT biased.
So let’s consider a specific case where a given healthcare algorithm, say, Algo33, trained on mountains of clinical data, turns out to diagnose the presence of Dreaded Disease #7 more reliably in white folk than in black folk. DD7 is a killer if not detected early, and in practice Algo33 saves thousands of lives each year, but proportionately fewer African-American lives. Algo33 is definitely biased. Should we then take Algo33 offline until it can be ‘repaired’ (not an easy task)? Or should we empanel a death committee to oversee Algo33’s race-based diagnoses, and adjust the white notification numbers so that they are equally deficient as the algo’s black notification numbers? Or come up with some other ‘compensatory approach’ that passes political muster? In such approaches we would be implementing regulatory policies based on the existential practice that ‘the perfect is the enemy of the good’.
rebanesruminations.com 
Leave a comment