Artificial Inteligence
rfp
sales
Some advocates of artificial intelligence (AI) declare that it is better at making decisions than humans because computers can’t be biased. Algorithms are based on math and logic, so emotions, prejudices, and predispositions are excluded. The AI engine is dispassionate, impartial, and neutral.
If only this was true.
A large language model (LLM) is built to make decisions based on training data. Initially, the program is fed scenarios where the correct response is already known. The data is labelled and the machine “is taught” the correct response. This is known as supervised learning.
This is also the first opportunity where human bias can be introduced. For example, if the training data reflects prior prejudicial decisions, then that becomes the basis for future decisions. This may include racial or gender discrimination, ageist conclusions, or unfair assumptions about disabilities.
These are usually the first thoughts when we read the word “bias”. However, there are many other types of bias that aren’t about societal norms. Biases may be based on economic models, managerial schools of thought, and accounting methodologies.
For example, for me, time is a precious resource. I approach most days thinking about what order of events will be the most productive. This even includes maximizing my leisure activities. My internal clock almost never turns off.
When training an AI engine, I may only include “correct” answers that favor timeliness over other factors. That could preemptively exclude solutions that take longer, but preserve other resources. The more times those answers are used, the deeper ingrained the “time bias” becomes.
Offsetting AI bias is a never-ending process. It requires rigorous assessments, improved data quality, and ensuring a diversity of perspectives across the design, evaluation, and deployment phases. Regular testing by multiple humans must be a continuous discipline.
Amazing Astronomical Fact: The Sun has reached its solar maximum period of its 11-year cycle, which will continue for the next year. That means additional magnetic activity and more sunspots.
With an average size larger than the planet Earth, sunspots are cooler regions caused by a concentration of magnetic activity.
The image at the with this post shows a band of sunspots across the face of the sun.
#artificialintelligence #ai #machinelearning #bias #sunspots #theberkshirecompany
Missed a past newsletter? No worries they're all saved on our Thursday Thoughts on AI page.
