“We don’t see things as they are, we see them as we are.” Thus wrote Anaïs Nin, describing rather succinctly the unfortunate bias mix that accompany our human brains, which otherwise function perfectly well.
In a business context, affinity bias, confirmation bias, attribution bias, and the halo effect, among the most well-known of these errors in reasoning, only scratch the surface. Overall, they leave a trail of infractions and errors in their wake.
Of course, the most pernicious of our human biases are those that prejudice us for or against our fellow human beings on the basis of age, race, sex, religion, or physical appearance. Try as we do to cleanse ourselves, our work environments and our society of these distortions, they still creep into – well, pretty much everything we think and do – even modern technologies, like the AI.
Critics say AI is making bias worse
Since AI was first deployed in hiring, loan approvals, insurance premium modeling, facial recognition, law enforcement and a constellation of other applications, critics have, with considerable justification, pointed to technology’s propensity for bias.
Representations of Google’s bidirectional encoder from Transformers (BERT), for example, is a leading natural language processing (NLP) model that developers can use to create their own AI. BERT was originally built using text from Wikipedia as its primary source. What’s wrong with that? The overwhelming majority of Wikipedia contributors are white men from Europe and North America. As a result, one of the most important sources of language-based AI started life with a biased perspective built in.
A similar problem was found in Computer Vision, another key area of AI development. Facial recognition datasets comprising hundreds of thousands of annotated faces are essential for the development of facial recognition applications used for cybersecurity, law enforcement and even customer service. It turned out, however, that the devs (presumably mostly middle-aged white males) subconsciously did a better job of achieving accuracy for people like them. Error rate for women, children, seniors and people of color were much higher than those of middle-aged white men. As a result, IBM, Amazon and Microsoft were forced to stop selling their facial recognition technology to law enforcement in 2020, lest these biases lead to the misidentification of suspects.
To learn more about it all, I encourage you to watch the important and sometimes chilling documentary Coded bias.
What if AI really was part of the bias solution?
A better understanding of the phenomenon of bias in AI, however, reveals that AI only exposes and amplifies implicit biases that already existed, but have been overlooked or misunderstood. The AI itself is independent of color, gender, age and other biases. It is not vulnerable to logical errors and cognitive biases that trouble humans. The only reason we see biases in AI is because of heuristic errors and biased data that humans sometimes train it with.
Since the discovery of the biases listed above (a PR disaster, I assure you), every major tech company has worked to improve the datasets and eliminate the biases. A way to eliminate bias in AI?—using AI! If that seems unlikely, read on.
Using AI to Eliminate Bias in Hiring
The classic example is found in job postings. Across all of the most coveted job opportunities, women and people of color are notoriously underrepresented. The phenomenon is perpetuated as new hires become senior managers and become hiring managers. Affinity bias ensures that “people like me” continue to be hired, while attribution bias justifies these choices based on past hire performance.
But when AI is given a bigger role in recruitment, that may change. tools like Text, gender decoderand ongig use AI to scrutinize job descriptions for hidden biases regarding gender and other characteristics. Knockri, Ceridianand gap jumpers Use AI to remove or ignore characteristics that identify gender, national origin, skin color, and age, so hiring managers can focus solely on candidate qualifications and experience. Some of these solutions also reduce recency bias, affinity bias, and gender bias from the interview process by evaluating candidates’ soft skills on an objective basis or by altering a candidate’s telephone voice to mask their sex.
Eliminating bias in venture capital decision-making with AI
A similar approach can be taken in the world of venture capital, where men 80% of partners and women, receive only 2.2% of all investments, despite being the founders of 40% of new startups. UK accelerator Founders Factory, for example, wrote software that screens candidates for the program based on identifiable entrepreneurial success characteristics. Similarly, F4capital, a non-profit organization run by women, has developed a “FICO score for startupswhich presents the maturity, opportunities and risks of startups as a way to eliminate bias in the decision-making process. This approach should be widely adopted not just because it’s the ethical thing to do, but because it offers better returns… up to 184% higher than investments made without the help of AI.
Reducing cognitive biases with AI in the medical field
AI can also help make better decisions in healthcare. Medical diagnostics company Flow Health, for example, has pledged to use AI to overcome cognitive biases which he says doctors often use to diagnose patients. The ‘availability heuristic’, for example, encourages physicians to make a common, but sometimes wrong, diagnosis, while the ‘anchoring heuristic’ leads them to hold to erroneous initial diagnoses, even when new information contradicts them. I believe that AI will become an essential part of the rapidly changing world of data-driven personalized medicine.
Other areas where AI can reduce common biases
AI can even help reduce the less malignant, but still very powerful biases that too often cloud our business judgment. Consider the bias (in English-speaking countries) towards information published in English versus other languages, the bias of the startup world against older people despite their greater knowledge and experience, and the bias in manufacturing to use the same providers and methods, rather than trying new and potentially better methods. Don’t forget the biases that drive supply chain management executives and Wall Street investors to make emotional, short-term decisions in tough economic times.
Giving AI a role in all of these areas is a useful check against unrecognized bias in your decision-making process.
AI can even be used to reduce bias in AI
If to err is human, AI may be the solution we need to avoid the costly and unethical outcomes of our hidden biases. But what about the intrusion of these biases into the AI itself? If AI misinterprets biased data and amplifies biased human heuristics, how can it be a useful solution?
There are now tools designed to eliminate the implicit human and data biases that are surreptitiously creeping into artificial intelligence. The simulation tooldeveloped by the Google People and AI Research (PAIR) team, allows developers to probe AI performance using a large library of “fairness indicators” while PWC Bias Analyzer ToolIBM Research AI Equity 360 tool and O’Reilly’s FILE tool each helps you identify if there is a bias in your AI code.
If you’re a senior executive or board member thinking about ways AI could reduce bias in your organization (it’s after all why I’m writing this article), I urge you to see AI as a promising new weapon. in your arsenal, not as a silver bullet that will solve the entire problem. From a holistic and practical perspective, you should always create bias reduction benchmarks, train your staff to recognize and avoid hidden biases, and gather outside feedback from customers, vendors, or consultants. Not only are biased audits a good idea, but in some cases, they are the law.
If you care about how AI determines winners and losers in business, and how you can leverage AI to benefit your organization, I encourage you to stay tuned. I write (almost) exclusively about how senior executives, board members, and other business leaders can effectively use AI. You can read past articles and be notified of new ones by clicking the “follow” button here.