Short Bytes: The resurgent debate on the impact of AI on humanity in the (near)future has drawn a lot of attention in recent times; Especially after the opposing remarks made by Elon Musk and Mark Zuckerberg. Here, we shall discuss the arguments and assertions put forth in the light of the current AI backdrop.
- “AI surpasses humans in Classifying Images” with 97.3% accuracy on the famous ImageNet data in 2017. [Link]
- AI suffuses texture and details to a roughly drawn image. [Link]
- AI transforms horses to zebras in a video. [Link]
- AI generates code to create the GUI, given as an image. [Link]
- AI makes robots learn from simulations.
These are some of the exciting news generated by the field of AI recently. These surely are noteworthy results that transcend the applications that we previously hadn’t thought about. These results indirectly expound the techniques and the mechanisms that were developed over the years. But the titles seem to have created an alarm about the future in the minds of the public. I implore the readers to check out the above links to get a sense of the amazing capabilities of AI, before proceeding.
We shall draw our attention to the general arguments made by people in this field, regarding the future. Leaving aside the notions of Artificial General Intelligence (AGI), the Kurzweilian plots and their surmisal inferences, we shall focus on our current concrete knowledge of AI.
The AI Overlord
This is the typical case of ethereal killer robot theory put forth by some people, including Stephen Hawking, with Elon Musk as their champion crusader. In Elon Musk’s words “(AI) poses a fundamental risk to the existence of human civilisation”. These ideas may have originated from the Sci-fi films, often portraying a bleak and an apocalyptic future whose cause is attributed to advanced technology and AI.
The idea of inhuman robots(pun intended) taking over humanity and ruling us all is terrifying. However, one surely cannot rule out this possibility. We have ample evidence of science gone horribly wrong.
Proactive measures are demanded by this group to regulate the research and development of AI and Robotics. Not only is this demand unnecessary in the current scenario, but it is implausible. Anyone with a good computing resource and a mathematical (and programming) knowledge can develop AI. With most of the AI research, its associated implementation and most importantly – the data required being openly published, regulation of it is almost impossible.
The AI community, including leading researchers, have widely condemned these as mere “fear mongering”. One should also keep in mind that Elon Musk founded the “Open AI”, a company dedicated to ‘enacting the path to safe AI’; and it is intriguing if at all those researchers educated Mr.Musk on these issues.
The AI Slave
The next notion put forth is that AI will displace human jobs. Machines with fast computing power and intelligent decision-making capabilities will replace humans in the near future. For example, with the advent of Algorithmic Trading, banks have cut the role of human traders by a large margin. The researchers, like Professor Andrew Ng, who openly deemed the previous theory as “doomsday-like”, whole heartedly agree on this issue.
This threat of economic dislocation is a far more serious problem compared to the previous one because the current AI is advanced enough to displace jobs. We can observe this happening already. Computers already displaced thousands on jobs in the past decade. With the ever-growing interest in AI, it is plausible that a significant fraction of them will develop machines to automate human jobs. It is, in a way, immoral to rob someone’s livelihood just because a student wanted to build a “Cool Machine” for his PhD.
People whose jobs border the AI research direction must be forewarned. But then, demanding the populace to know of the developments in AI and its immediate impact is highly burdensome. It is the researchers’ duty to see that people are well aware of the implications of their research. Even then, it does not heal the wound, does it?
Ponderings and Observations
Most of the news generated by AI predominantly focus on images and its associated tasks like generation and classification. Furthermore, the hardware developments seem to tend towards this too. But then, in the previous decade, data classification and explorative robotics were in focus. In the next decade or so, the focus will shift more towards human language. Each decade in AI seems to conquer one particular field. This forms the basis for the debate that, how long before AI conquers everything that is Human.
One of the recent trends is that most of the AI companies and research institutes have started openly blogging about their research results. Some of these blog posts are often accused of over hyping their research for PR purposes. These, added with the exaggerations of the media, result in bombastic titles that inflict fear (or awe) in the minds of the general public.
The ultimate question is, can AI ever be advanced enough to compete with Human Intelligence? The answer is – specifically yes, in general – mostly NO!. Consider this plot, from the Deep Learning book by Ian Goodfellow et. al. (considered to be one of the best books on the subject)
The numbered red boxes indicate the different architectures developed over the years (x-axis), whose complexity (y-axis) is compared to that of animal brains. On the surface, it gives an alarming impression that we are not far from developing an AI architecture that is on-par with the complexity of human brain. Unfortunately (or Fortunately in this case), complexity does not directly translate to capability. The architectures that are more complex than the brain of a mouse cannot do everything that a mouse can; even though they might be better at specific tasks.
It is left to the reader’s supposition whether we will create an AI that is on-par with human intelligence on holistic tasks. But a lot of research is currently being oriented towards this direction. However, believing the collective wisdom, it will take a long time for that to happen, if at all it does.
It is kind of recursive to think that with the advance in AI, humans will be forced to work on AI itself (because those might the only jobs left), which again displaces more jobs.
One solution is to leverage the vast expanse of the diverse economy that we have created, to invent more jobs. 20 years ago, data analyst, IT professional, front-end developer, etc. were unheard of. Similarly, in the future, humans will come up with jobs and take on roles that we might not be able to even speculate now.
The other, more trending, solution is to integrate humans and machines at work; considering machines as collaborators in solving problems and getting work done. AI is not going away anytime soon and we have to make peace with the fact that it will change our lives profoundly.
I will leave the philosophical question of “what makes us human if machines can do everything that we do?” to the reader’s rumination. But one thing is for sure – if at all we wind up at the desperation of identity crisis, we will redefine ourselves… yet again.
Cover Illustration by Paul Lachine.