Folks,
You have undoubtedly heard in the news that some predict AI will destroy civilization. Even the “Godfather” of AI, Geoffrey Hinton has quit Google and now feels compelled to warn of AI’s risks.
I recently spoke on this topic at Bio 2023 in Boston. I don’t want to minimize the stunning advances in AI as displayed recently in ChatGPT. But, its threat to civilization is at worst nuanced.
However, those of us in the teaching world are concerned of the impact of ChatGPT in education. It can write quite good papers on any topic, although errors are common. However, its success in tackling exams is nothing short of stunning. ChatGPT can pass the bar exam for lawyers at the 90 percentile, ChatGPT4 can also pass a wide variety of standardized tests such as SATs, APs, GRE’s and even a test for Sommeliers! These skills of AIs like ChatGPT raises the concerns of academic cheating and the hindering the learning of our youth.
An article by the New York Post sums up these types of concerns: “Long known for his warnings on the potential dangers of AI." Tesla CEO Elon Musk on Monday cautioned that even a “benign dependency” on these complex machines can threaten civilization.” (NY Post May 2, 2023)
Another concern RE AI is crime. Criminals are very clever and will use AI to scam people among other things. One way scamming is done is through deepfakes. A typical deepfake is a digital file of a person’s voice or video image that is fake. There is no end of ways that criminals can use deepfakes to extort money.
So, the latest AI technology is indeed unsettling. However, there is one area in which AIs are utter failures and likely will be for generations: The Physical Embodied Turing Test. This test would essentially challenge an AI robot to assemble something like IKEA furniture from a kit with written instructions. The state of AI development is so far behind in this regime that it makes this task science fictional. In 2019, my then 8 year-old grandson, Nate Su, assembled the Apollo Spacecraft’s 1969 parts, see Figure 1, in 4 hours. No AI can come close to doing this. For AIs to be a real physical threat to humanity, significant advances would be required in interacting in the physical world.
Figure 1. Eight year old Nate Su assembled the 1969 parts of the Apollo rocket in 4 hours. No AI can come close to this feat.
One of the reasons AIs performs so poorly in the physical world is that they have no body. So much of human’s interaction with the physical world is through our bodies and our senses of taste, touch, sight and sound. In addition, humans have years of context about the physical world that an AI Robot would have to learn. As an example, as I write this post I can look out into my living room and see the hard wood floor, covered by carpets, the furniture, our Mumford fireplace, the TV, the phone, and on and on. Each of these items have a long story connected with them. For an AI to navigate our physical world, it would need to “learn” about the myriad aspects of it.
So, don’t have any concern that an AI robot like M3gan (Figure 2.) is on the horizon. Even more humbling to the AI world is Steven Pinker’s adage that no AI can empty a dishwasher.
Figure 2. Robots like Megan will not be on the horizon for decades, if ever.
Cheers,
Dr. Ron