AI, you are what you eat.
It is no surprise that “Norman”, the world’s first “psychopathic” artificial intelligence (AI), could be created and act like a lunatic. Massachusetts Institute of Technology (MIT) researchers fed it data from sick and violent Reddit forums then tested it by comparing it to captions from a standard image captioning neural network using Rorschach’s inkblot tests.
You are what you eat; aka, GIGO — garbage in, garbage out.
The implications are profound for at least 2 reasons:
- In the near future, anyone can make an AI and many will; most will have an agenda or a skewed view of “good” data.
- Few, if any, will know what data was eaten; everything is opaque so the black box just gives you an “authoritative” answer.
There will be artificial intelligence of all races, cultures, religions, political and financial orientation. And you as the inquisitor, will not know why you are getting a particular answer.
Consider the broad concepts of creationism vs. evolution in a literal interpretation vs. science discovery argument.
Let’s feed the AI with the body of knowledge (a corpus of information) built from the billions of words written in favor of creationism and the literal interpretation of the Bible. Let’s also add billions of words that align with it from Evangelical teachings around the globe.
Then start asking this “Creationist AI” a question: when was the earth created? Answer: the world was created between 5,700 and 10,000 years ago. In part this is because the AI attributes importance to a Gallup survey that reports 38% of USA adults answered that “God created humans in their present form at one time within the last 10,000 years.”
Not to pick on religion as an easy target, but controversy is pervasive in the “sciences” of medicine and nutrition. Opinions about sugar and high fructose corn syrup have raged on for decades as study contradicts study. Same for medical issues around cholesterol and high blood pressure, their causes and treatment, using diet vs. pharmaceuticals are still subject to divergent facts.
Let’s not even get into the voodoo of vitamins and cosmetics, and the how or why organics or minerals matter.
Also in terms of medicine, countries around the world differ on what best practice treatment might be recommended, such as herbals over drugs or acupuncture over surgery, so an expert medical AI in China could yield different answers that may or not be correct to whomever.
While best intentions might be clear for many AI, the reality is commercial, advertising and sales oriented AI will invade the environment. What happens when you ask the Amazon Alexa or Google Now “what is the best <item>?” How will you know the accuracy or reliability of the answer?
For example:
- Question: AI, where is a popular beach destination in January?
- Answer: The Cancun <brand> resort is popular and has a special offer that includes Airline discounts and <brand> meal vouchers.
- Question: AI, I need more calcium in my diet.
- Answer: Milk is popular but <brand> cheese is more tasty.
With apologies to Forrest Gump, “stupid is what stupid eats.” And with AI, the data is the artificial ingredient.
Ask me anything. Click to read more about AI and branding.
Larry Smith
linkedin.com/in/lwsmith10011
