Nine AI myths versus reality
Are killer robots coming to take our jobs? We debunk the top misconceptions about artificial intelligence
Whenever artificial intelligence, aka AI, comes up in conversation, the usual image that springs to most people's minds is a threatening killer robot along the lines of Terminator that has nothing but murderous intentions towards humanity.
But these days the AI acronym is being liberally sprinkled far and wide, often referring to things that stretch well beyond its primary meaning. In this feature, we look at some of the common myths and misconceptions, compared to the real scientific situation in each case.
1. AI will create a malevolent Skynet-style system that will destroy humanity
Let's look at the most popular myth first the scary robot elephant in the room. Terminator is the most well-known example, but it is a recurring sci-fi theme from 2001: A Space Odyssey to the latest season of Star Trek: Discovery. On the one hand, technology has been automating the delivery of ordnance for decades, with in-missile video footage from the 1991 Gulf War just one watershed moment in a process towards greater autonomy that dates back to the German V1 and V2 rockets of World War II. The US army has been deploying Unmanned Air Vehicles (UAVs) for decades, and now has around 10,000 of them in regular use. But all of these still have human operators for key functions. The Defence Advanced Research Projects Agency (DARPA) has been awarding grants for the development of UAVs that can navigate themselves indoors. But as the RAND corporation points out, very few countries use armed drones just yet, and there is much controversy about their central value in warfare compared to conventional weapons systems. So even if fully autonomous fighting machines are developed, there are still many hurdles before they are deployed without human oversight, let alone take over the world.
2. AI systems and robots will eventually replace all jobs, making most people redundant
According to a report published by the UK's Department of Work and Pensions, 8,820,545 jobs could be wiped out by 2030 because of AI, particularly in the retail sector. Aside from the strangely specific number of job losses predicted, it's worth noting that it won't just be menial labour that gets replaced. Complex intellectual activities are already being replicated by expert systems, such as legal and medical advice. AI has been making inroads into healthcare to allow earlier diagnosis without the need for consultation with specialists, who are always at a premium. You can even put your job title into the Will Robots Take My Job website to see how likely you are to be replaced by AI. In reality, though, similar arguments have been made since the agrarian and industrial revolutions. On the one hand, many jobs will be automated by AI, but on the other, people can retrain, or young people educated in a different direction for the new jobs that are emerging potentially designing and building those AI-powered robots.
3. Siri, Google Home, Cortana and Alexa are AI
Voice-activated speakers have been a Christmas hit for the last couple of years, and more people are getting used to giving the smartphones verbal commands, too. These are undeniably clever, convenient systems (when they work), but in reality, they are just advanced natural language processing (NLP) recognition algorithms akin to dictation software like Dragon Naturally Speaking. There is no original thought going on, just a lot of pre-programmed responses to verbal commands.
4. AI is a computer version of the human brain
We now get to the main underlying myth of AI that computers model the human brain. This could be the subject of multiple PhD dissertations, but in a nutshell (and just for starters), computers are still based on the Von Neuman machine model of the mid-1940s. This reads data from memory, operates on it, and writes the result back to memory. This is not how brains work. Even multi-core processors, or HPC datacentres full of them, are still much more serial in their operation than a brain, which in contrast has a slow frequency (around 200Hz compared to multiple Gigahertz) but is massively parallel. Not just massively parallel, but inputs and outputs are connected in complicated feedback loops, with workings that we still don't understand completely yet. This isn't to say that computer AI isn't amazingly useful, or that we won't ever fully understand the human brain. But current AI is at best a very rough simulation, not even close to a digital facsimile of the cerebrum of homo sapiens.
5. AI systems can learn for themselves
Another two-letter acronym often found alongside AI is ML (Machine Learning). The common myth is that ML is a fully autonomous process, which will potentially lead to AI that transcends human intelligence and eventually decides to get rid of us (see 1 above). However, ML still intrinsically involves teaching by humans. Every AI system needs to be fed source material chosen by people, and its outputs adjusted by human experts until they work. This is precisely the process that Google's self-driving car system is going through right now, and this won't stop even when it gets the green light as a commercially available system in new vehicles.
6. AI systems will be much more impartial than human beings
As a result of AI's ML being fed by humans, there's no reason to believe that it will be any more impervious to prejudices than the humans that taught it. Early facial recognition systems had trouble identifying ethnicities, and the Tay Twitter bot was rapidly turned into a rabid racist by the tone of social media conversations. On the other hand, an AI that has been trained to be as impartial as the best human examples will consistently be better in this respect than the worst humans, which is where there is clear value for the legal and medical professions, amongst others.
7. AI will soon be smarter than human beings, or never will
Because AI runs on computers that are not an exact replica of the human brain (see four above), this is a bit of a trick question. On the one hand, there is no sign that a general AI will transcend overall human intelligence anytime soon, because we still don't know exactly what the latter is. But on the other hand, much more narrow AI has been beating humans for some years now, such as defeating the best human chess player, and surpassing the best players of Jeopardy!. In 2004, none of the vehicles in the DARPA Grand Challenges completed the course, but in 2005, five did, and now we have self-driving cars being tested on public roads. So AIs will likely be smarter than humans in many key areas (and already are in some), but may never be in a general, overall sense whatever that even means.
8. The technological singularity is approaching, or will never happen
Related to seven above, there is a theory, originally presented by maths and computer science professor Vernor Vinge, that the development of artificial superintelligence will arrive around 2030 and then AI will upgrade itself beyond human understanding and the world will change unrecognisably. This has been dubbed "the singularity". For reasons already discussed above, the date already looks massively optimistic before we even get into the details, due to our lack of understanding of the human brain. But, on the other hand, the singularity doesn't necessarily have to involve completely human intelligence. We could be creating something that isn't human, just loosely based on us. Nevertheless, this idea still gives computers the ability to have their own intentions, which somehow emerge spontaneously from the advanced technology. Right now, no AI can do anything other than what humans told it to do, even if your flaky desktop PC might sometimes make you believe otherwise.
9. AI is just sci-fi and nothing that should concern business
Just as AI can create new jobs as well as replace old ones, it's a topic that should be central to all businesses that intend to survive and grow over the coming decades, rather than ignored. On the one hand, the scary sci-fi scenarios of human replacement are very distant if ever likely to happen at all. But, on the other hand, there are real opportunities to enhance business services and consumer interaction via the more limited systems that have been shown to be effective already. Business intelligence data analytics, forward resource planning, and automated customer service are just the beginning. Companies that embrace AI, whilst being realistic about its limitations, will be the ones that prosper.