Can machines learn morality?

Humans have been using technology as a tool for a very long time, and there are many things the human brain still can’t do. But our brains have also evolved technologies that have allowed humans to self-advise, to learn without input. Is it possible that machines will also one day be able to learn morality?

In the 1960s and 1970s, artificial intelligence researcher Marvin Minsky came up with a way to teach machines to produce aesthetically pleasing images. As Minsky explains, the project, called M Ayer, used what is known as “dome logic” to teach the computer to generate images from written descriptions. For the project, Minsky taught a special computer model named Ayan (both words were inspired by Filipino archbishops Ayan and Migan) to pick an image from a gigantic screen, a program generated each of its images and then turned each picture over to another program. This program checked on the Ayan over the course of several minutes to make sure it had rendered the image correctly. If an image needed a little editing, the Ayan program would correct for that, with the resulting image uploaded onto the server and then publicly available to the world. Ayan taught itself to do all of this, and Minsky decided to use this technology to teach people to draw.

Over time, it became clear that people liked drawing pictures with a computer, and Minsky built on his original research and created a program called The Woodborrower, which employs “dome logic” to allow people to draw pictures. The program generates an image from a written description, which then tells the person how to draw that image, and the program then uploads the drawing onto the file hosting site. No human-to-human interaction, no learning, no instruction, no trial and error. In a fascinating 2015 TED Talk, the now 80-year-old Minsky explains that “a computer can perform the same tasks as an artist,” and that “all of us can learn to draw.” This is a technology that could one day become more widespread: Minsky reported that people in the original research group started drawing very quickly.

Many of us will never be able to draw very well, but it is already possible to test out this technology. If you use a public document — such as the press releases for a new antivirus program or an earnings release for a public company — you can upload those files to a server, and in a matter of minutes the computer will spit out a painting or a movie. There are plenty of recordings that provide instructions, such as this excellent tutorial from Google, but if you really want to make a painting, there is no one-stop shop for creating art. Thanks to Minsky’s brainchild, there is now an easy way for me to have a computer create something beautiful.

It’s certainly not the first time that computers have become friendly if they learn to express their own emotions. Machines have been able to read emotions for quite some time. The business side of Google’s search engine was powered by an AI who, starting in 2004, “translated every query into more than 200 natural-language expressions” that would then be displayed to users in order to guide them through the process of finding exactly what they were looking for. The website that provided this new service? Google. This operation, called A.I. Translation, let the computer learn what it meant to be happy, upset, conflicted, worried, and confused, and then trigger their responses. The idea of both humans and machines making things more machine-friendly was deeply ingrained in the company’s DNA. The company also built a team that taught the computer how to absorb facts and then translate them into natural language.

The takeaway here is that machines can learn from the wisdom of the crowds. In other words, computers can make delicious food, compose a beautiful masterpiece or aid a refugee.

It remains to be seen how far this technology will go, and some critics doubt that even the most loving A.I. will be able to learn morality. Minsky points out that human sense of fair play is linked to our moral development, and that people learn to hate, to help, and sometimes even to be indifferent to pretty much anyone and everything at some point in their life. A.I. learning to forgive, A.I. learning to help, A.I. learning to prefer certain emotions, it’s a tall order to ask. What if our machines fail to learn? And we have destroyed our capacity to be empathetic? What if humans are fundamentally better than machines? What if we use technology as a tool?

We have previously considered machine translation and R (for Real Time) messaging systems, but

Leave a Comment