Students, faculty and community members interested in how artificial intelligence operates and the impact it could have gathered at the CUB Orndorff theater on March 4, to hear Professor Alice Armstrong’s overview from the Department of Computer Science. 

Armstrong discussed the way AI learns, and the number of resources required to run the intricate system. A plethora of concerns are rooted at the core of AI’s ability to provide correct answers. Along with that is the devastating impact on the environment and resources to allow its operation to be possible, according to Armstrong.

AI uses a large language model to give it the ability to learn. It holds a loose connection to how neurons in our brains work, but without the knowledge to turn experiences into meaning like people do. Instead of having the different sets of feedback we get through our senses, a computer will simply figure out what should be next in the pattern of information it has been trained on. 

With the copious options that stem from having probabilities running through large sets of possibilities, even the most likely outcome is very unlikely in the broad scheme of things. 

AI can be great at taking lots of data for a specific topic and generating what outcomes may be next, but it will easily lose the plot of its objective if given too much wiggle room outside of the key idea one is trying to generate information on, according to Armstrong. 

A Columbia Journalism Review (CJR) study found that chatbots can be wrong up to 60% of the time, but the levels at which AI can make incorrect assumptions pale in comparison to the number of resources needed to run data centers.

Armstrong stressed that understanding how much water and electricity are needed to let AI function cannot be overlooked. With some models requiring as much energy as whole nations annually. 

Armstrong said an estimate from last year stated that generative AI models consume about 0.5% of all the world’s electricity. Generating a single AI image uses about as much electricity as charging a cell phone. 

Armstrong herself switched to DuckDuckGo, an internet search engine that will allow her to turn off the AI assistant that Google constantly has running, so that it is not used haphazardly. 

“Large data centers can consume up to 5 million gallons of water per day per center,” Armstrong said. “Including the fact that the water used must be fresh water.” 

While AI may become more efficient in the future, that is a long way off and is not critical work being invested in by the companies running the data centers, according to Armstong. 

AI can be useful in specific fields, but it is crucial to ask whether it matters if the output AI gives is true, if you can verify the accuracy of the answers given and if individuals are willing to take responsibility for inaccuracies. 

Armstrong’s final note was this: use AI with care and caution. 

Share