AI is making progress, but it is unlikely that this will succeed quickly in one important area
Comment: artificial intelligence can be in important areas such as perception, it is terrible in predicting social outcomes, says a professor from Princeton.
Image: Getty Images / iStockphoto
It will take time, but at some point every application will have its share in “AI Inside”. Today, however, we are far from that point, and false advertising for AI opportunities does not help, has something that Arvind Narayanan, associate professor of Computer Science at Princeton, has called “snake oil” in a recent presentation. It is not that there are no real, usable ways to use AI today, but emphasizes that “much of what is sold today as” AI “is snake oil – it doesn’t work and can’t.”
Where does Narayanan think we are making real progress on AI, and where should we break the myth to help dissect good and bad AI ads?
Be real about AI
As with any new technology, ambitions to embrace it always surpass actual production use, and AI is no different. According to a Gartner survey published earlier in 2019, 59% of companies surveyed today use AI and of those 59% they have deployed an average of four AI / ML projects. Gartner estimates that the average number of AI / ML projects deployed in 2020 will almost triple to 10, double to 20 in 2021 and reach 35 in 2022. “We see a significant acceleration in AI approval this year,” Jim Hare said, research vice president at Gartner.
SEE: Special report: management of AI and ML in the company (ZDNet) | Download the free PDF version (TechRepublic)
According to the same research, organizations tend to use AI / ML in the area of customer experience (supporting decision making and making recommendations to employees, such as offering almost real-time data to customer service representatives) and task automation (eg, invoicing and contract validation) in finance). These are reasonable ways to use AI, according to Narayanan.
Less reasonable are survey responses suggesting that 54% of the general population believe that AI will be able to “perform almost all tasks that are economically relevant today better than the median (today) for each.” As Narayanan pointed out, “AI experts have a more modest estimate that artificial general intelligence or strong AI is about 50 years away, but history tells us that even experts tend to be wildly optimistic about AI predictions.”
According to Narayanan, there are two key areas in which AI performs well today, the first being “Perception”, a category under which it falls:
-
Content identification (Shazam, reverse image search)
-
Face recognition
-
Medical diagnosis of scans
-
Speech to text
-
Deepfakes
In the area of Perception, Narayanan said, “AI is already at or higher than human accuracy” in the areas identified above (e.g., Content Identification) and “is getting better and faster.” The reason that it keeps getting better, he emphasized, is simple:
The fundamental reason for progress is that there is no uncertainty or ambiguity in these tasks – given two images of faces, there is fundamental truth about whether they represent the same person or not. Thus, if there is sufficient data and calculations, AI learns the patterns that distinguish one face from the other. There have been some notable face recognition failures, but I can safely predict that this will be much more accurate .
Of course, he noted, it is precisely that accuracy that means we have to be careful how it is used.
SEE: Guide for IT leaders for in-depth learning (TechRepublic Premium)
The second area where Narayanan indicates AI is performing well, although not as well as Perception, is automating judgment, including:
As he stated: “People have a little heuristic in our mind, such as what is spam and not spam, and given enough examples, the machine is trying to learn it. AI will never be perfect in these tasks because they judge and reasonable people can’t agree on the right decision. “AI will continue to improve in such areas, although we need to figure out the right procedures for correcting machine-driven decisions that are too far from human judgment.
AI is imperfect in these two areas and is becoming increasingly useful. But when it comes to predicting social outcomes, Narayanan is flourishing, the role of AI is “fundamentally questionable.”
Put AI back in place
In such areas, where ethical concerns are bundled with accuracy, AI today is not only a poor predictor, but it is unlikely that it will improve quickly. Examples of this are:
-
Prediction of criminal recidivism
-
Predict work performance
-
Predictive police work
-
Predicting terrorist risk
-
Predicting children at risk
Nor is it a matter of throwing more data at the problem. Using an example of predicting child outcomes based on 13,000 family characteristics, Narayanan complained that “AI was (better) hardly better than a simple linear formula” that used only four characteristics. Manual scoring, he continues, works better for predicting outcomes.
Moreover, when we rely on pseudo-AI for predicting social outcomes, we come across the problem of accountability (or rather the inability to explain the prediction): “Imagine a system in which, instead of points on a driver’s license once you are persuaded, the policeman enters your data into a computer. Most times you get free, but at some point the black box system tells you that you are no longer allowed to drive. “Without explaining why, we would have a new one to enter an even more destructive era of anger.
Again, this does not mean that AI is not a powerful force for society – it is, and AI will eventually find its way into almost any application. This is very good. It only gets bad if we misuse AI to predict social outcomes, in Narayanan’s thought, without the backstop of even being able to explain the employee, potential terrorist, etc. why they are fired, arrested or arrested. worse.
revelation: I work for AWS, which has AI / ML related products. However, I do not work for these product teams and nothing herein is intended to promote or refer to any AWS technology.
Next Big Thing newsletter
Be up to date with smart cities, AI, Internet of Things, VR, autonomous driving, drones, robotics and more of the coolest technical innovations.
Delivered on Wednesday and Friday
Register today
