08/01/2019
What causes bias in algorithms?
AI is only as good as the data used to create it. Too little data or incorrectly labeled data leads to inaccuracy in algorithm performance.
Data is where humans come into the picture. Data cannot be labeled automatically by a computer; a human must manually mark each piece of information.
"The position humans have in the AI stack is frequently misunderstood… People play a critical role in [creating algorithms and organizing extensive datasets to bring AI to life], so developers must build training and safeguards into their process to identify and reduce bias in AI systems. Since many would-be disruptors source at least some of their data annotation to third parties, this concept extends to their vendors and service providers.”
So, what's is the problem?
Scrolling through and marking each piece of data with a label is tedious. And AI needs TONS of data. Many AI companies outsource this part to huge labeling services, such as Amazon Mechanical Turk. Workers are paid per image (not for the quality of labels) leading to carelessness and bias in labeling. Moreover, AI companies have no idea who is labeling their data.
Who is responsible, then?
The answer to this question is still unclear.
There is light at the end of the tunnel, however:
AI can be unbiased. Unbiased data reflects reality. To take control of bias at the root cause, we use our own labeling team. We watch over labeling rules. We are diligent with corner cases. We work face-to-face. We combat bias in AI and in doing so, bring a little more truth into the world.
The fight with bias in AI is ongoing and will persist. What do you think AI companies should be doing about bias?
As long as people are developing AI technologies, bias is likely going to be a lingering issue. Similar to Data in Star Trek, the endeavor to reduce bias will be an evolution that requires us to strive for greatness
