Machines will Never Think Like Us

Machines will Never Think Like Us
(nor will we want them to)
(c) 2018 J.D. Chapman — (All Rights Reserved)

Executive Summary

Robots with generalized learning capabilities (so called AI) trained to specific objectives are commonly being deployed today. This essay argues that as such devices become widespread and commercially available they will mainly gain popularity for use to perform non-risky household chores, or as convenient rental friends for practice when a human partner is unavailable. Due to the intricacy of culture, intelligent robots are unlikely to be useful or targeted for a wide range of human activities, and hence will never replicate our thinking patterns. (8 pages)

***

Although we’re still at the early stages of Machine Learning, folks are getting their underwear in a twist about the prospect of being replaced by “intelligent” machines [1-3]. In this essay I will demonstrate why this is unlikely, as machines will never think like us (nor will we want them to). I will suggest how robots possessing human-like learning capabilities might be helpful.

The terminology used to refer to current advances in machine learning is still evolving. Popular press calls it AI (artificial intelligence) but this turn of phrase implies a synthetic replication of a human trait, “intelligence.” This is not however a driving goal of current research or technology [4]. Researches are trying to leverage models of neural connectivity to create useful algorithms and machines, yet there is little “artificial” about it, nor do such machinations strive for “intelligence.” Developers gear them toward specific measurable ends, as tools that humans may utilize and learn from [5].

This achievement can be better referred to as Machine Learning. Software systems trained to specific objectives, such as playing Chess, assembling an automobile, or facial recognition, are commonly being deployed today. Systems geared toward more generalized tasks (such as IBM’s Watson) are still very much under development.

Some in the field refer to such learning systems as Artificial General Intelligence, or AGI. In this essay I refer to more robust generalized learning machines as Natural Learning Machines, or NLMs. Conceptually this is software and supporting hardware that has both the potential and the experience to learn useful applicable knowledge in a self-directed fashion across a wide range of topics. If the hardware is additionally capable of robotic activity to physically move objects about in the real world, this essay refers to them as NLM Robots, or NLMRs.

Technology does not progress in a vacuum, and to properly emplace NLMRs in the future requires context for how other technologies are likely to mature when NLMRs become widely commercially available. This essayist believes a couple of other technologies will ripen first, namely widespread use of drones, and the popular adoption of AR/VR glasses (or contact lenses). Hence we will choose to train NLMRs upon those tasks not already provided by those technologies conveniently available.

The best way I can elucidate why NLMRs won’t think like us is to present a thought experiment that turns out to be preposterous prima facie. Can you envision a future where NLMRs will gleefully attend MLB baseball games on their own? You know, just to enjoy the Dodger dogs, 18 dollar beers, camaraderie of “doing the wave,” and the joy of clapping their hands and whistling at players? No, you cannot — this will never happen, and I will explain why.

For one, it is of no benefit to them or to anybody else (unless the MLB finagles a way to lure them in for the price of a discounted baseball ticket and some marketing research). If you personally owned an NLMR though and happened to be interested in a baseball game, you could just as easily turn on the television or stream the game on the internet in full Virtual Reality.

Moreover, a lot of the joy from attending sporting events is the shared cultural history of those events that would be incomprehensible to any NLMRs. We don’t just watch a batter attempt to hit the pitcher’s best curveball — long time fans see the complete context of how the pitcher, batter, and teams in this situation have performed this role before. Baseball players have their own quirks and personalities that aficionados follow, and the sport is prone to its gaffes and surprises. When these coincide you achieve a collective fan convulsion. For example consider one frontier of pitching personalities: Bartolo Colon, nicknamed Big Sexy. As with most pitchers, he’s a remarkably mediocre batter. This year when he hit a home run [6], fans (and all of baseball Twitter) was aghast. He became the quarry for a slew of “surprise look” memes.

All of this is well beyond anything an NLMR would even remotely consider as worthwhile, yet this camaraderie is at the heart of much human business and activity. It would be pointless to teach NLMRs any of this, as the underlying effect is a desire for /human/ entertainment on terms related to the specific values of that sport. For every business that deals with aesthetics or entertainment (together a considerable chunk of the economy) the target audience is, and always will be, humans. More specifically humans within a certain cultural background.

Excluding this rather large sphere of activity from the realm of NLMRs, what might we wish of them? We would best utilize them where they could perform work that could be complicated or stressful for us ourselves, provided that the outcome is not geared toward culturally aesthetic ends, nor situations that might be potentially risky [7].

I can imagine quite a mountain of tasks where it would be handy to have an NLMR standing by. Say I was going to make some onion bread. “Hey Robot, chop up this onion for me please.” Great. “Hey now can you take out the teflon saucepan and brown that onion in some butter with a tablespoon of canola oil?” Awesome. Just let it cool on the stove when you’re done bro.

“Hey Robot, the place needs a good dusting, and vacuum afterwards please. Can you also do a load of white laundry for me?” “Hey robot, pack my gym bag please, with the gray fit-dry shirt, and leave it by the front door.” “Hey robot can you clean my work shoes?” “Robot I left the extra battery for my laptop in my car trunk — can you go retrieve it and then plug it in to recharge?” “Hey robot, can you go clean the shower for me?”

Seemingly menial chores lacking panache, these tasks contrast with the realm of how companies market intelligent robots today (say toward mechanized guard dogs or military autonomous weapons [8]). But not so unlike most consumer technology, the real money and wealth generation will spring less from the initial government-sponsored uses as how these NLMRs eventually facilitate the exigencies of the common man. And all of the above examples are cases where I could defray some personal hassle by relying upon an NLMR without much of a possible downside.

I grant you that what I might ask of an NLMR may be totally different than what you might ask. You may have yours wash the dishes, but I won’t, as I rather enjoy that task myself (it has a certain Zen relaxing feel to it). Or you may have the NLMR knead a loaf of bread for you, but not me, as I find this activity to be manually therapeutic. I expect a broad diversity in the tasks we ask of our robots. Even as individuals we may sometimes choose to have the NLMR perform a task and sometimes not. If I’m too tired right now and the sink is full, then okay “Robot please wash, rinse, and stack the dishes into the drying rack.” Thanks.

Initially, robots with NLM capabilities will be quite expensive [9], with a significant amount of sunk R&D costs even beyond that already funded by the military. Hence they are likely to primarily be shared resources: in a medium size apartment building you might have one community NLMR available for a couple of hours each day. There might be one wandering around in your workplace.

NLMRs will also gain popularity for use as rental “practice friends,” in situations where you would normally prefer to do something with somebody else, but it’s more convenient at the moment to use the robot. Maybe you want a little 3-on-3 basketball but you’re short a couple players. Or your tennis instructor can’t make it today. Maybe your favorite manicurist is out of town.

I have read a few articles about NLMRs infringing on the realm of artisans [10]; whereas I can envision them as a useful tool in a generative sense, I don’t foresee them ever acquiring the finely attuned sense of culture or temporal fashion to be successful on their own. A human artist can, however, guide them in a lucrative joint creative effort.

Once you realize how they will likely be utilized, you can appreciate that we’re not talking about C-3PO, T-800 Terminators, RoboCops, Transmorphers, or Blade Runner replicant types of intelligent robots at all. Our NLMRs won’t have political agendas of their own, they won’t care about what television shows get cancelled, they won’t get angry or depressed and retain a psychiatrist, they won’t join sororities or gather swag from trade show conferences. They won’t go to museums of art.

The abundance of commercially available intelligent robots of the future will be personal task assistants and rental practice partners. A large cadre of human technicians will manage, maintain, upgrade, and certify these NLMRs. Much in parallel to how automotive manufacturers and dealerships operate today, a whole industry will evolve for the marketing and advertising of their capabilities, and their subsequent sale, enhancement, maintenance, customization, rental, and resale. Lobbyists will press governments for favored status of NLMRs best meeting their own objectives, and conscientious organizations will either protest or promote NLMR usage based upon their own agendas.

Most certainly though, intelligent robots will never think like us.
—————–

[1] https://www.salon.com/2018/01/06/what-an-artificial-intelligence-researcher-fears-about-ai_partner/
[2] https://www.cnbc.com/2018/01/04/south-koreas-lg-electronics-to-introduce-new-robots-at-ces-2018.html
[3] https://www.theguardian.com/business/2018/apr/03/robots-could-take-over-more-than-65m-jobs-warns-oecd-report
[4] http://saiconference.com/Conferences/IntelliSys2017
[5] http://newvantage.com/wp-content/uploads/2018/01/Big-Data-Executive-Survey-2018-Findings-1.pdf
[6] https://www.mlb.com/news/bartolo-colon-hits-first-home-run-of-career/c-176830856
[7] https://www.techradar.com/news/the-best-robots-of-ces-2018
[8] https://www.researchbeam.com/global-military-robots-2017-2021-market
[9] https://www.theverge.com/2018/1/10/16865506/laundroid-laundry-folding-machine-foldimate-ces-2018
[10] https://creators.vice.com/en_uk/article/qkw4vq/will-robots-replace-artists-in-the-future