We are living in an age where artificial intelligence is being channelized or even 'advertised' as being channelized in realms where only the imagination could once reach. The speed of AI seems to be outpacing Moore's Law. A Stanford Report states "Prior to 2012, AI results closely tracked Moore's Law, with compute doubling every two years. Post-2012, compute has been doubling every 3.4 months." With such a breakneck speed of advancements taking place in AI, one cannot help but ask questions such as: Is it all happening for the larger good? Is unconstrained growth the best way forward? Well, as noted speaker and marketing professor Scott Galloway says, "Nothing is ever as good or as bad as it seems."

I am of the camp that firmly believes that the energies being focused on AI advancements will benefit more than harm and may even provide us a fair shot at addressing bigger challenges such as climate change and defeating cancer. But for that to happen, AI needs to be governed, regulated, and channelized based on principles laid down by human minds who have the benefit of hindsight, empathy, and rationale – traits which still elude AI. This is where 'responsible AI' has an enormous role to play.

What is 'Responsible AI'?

Responsible AI is a framework which enables AI solutions to focus on the problem while not losing sight of critical aspects such as fairness, reliability, privacy, explainability, and inclusiveness. Without these dimensions being addressed during the development of any AI solution, it will not be sustainable and would not meet the expectations for adoption. Now, let us look at these tenets individually.

Fairness – AI solutions should ensure fairness towards all segments of the target population, in the data used. Unfairness owing to data limitations should not reflect in the eventual solutions and should be addressed and eliminated if present. As a simple example, consider the term 'Software Engineer'. It is gender-neutral and hence care should be taken to ensure that its use is neutral and equidistant from genders.

Reliability – Simply put, AI solutions should be trustworthy, consistent in outcomes, and prove reliable in performance across various use cases.

Privacy – AI solutions should perform within the generally accepted global guidelines of data privacy. Privacy rules should never be flouted in the pursuit of accuracy. What's more, solutions that give importance to data privacy are more trustworthy, reliable, and lead to more sustainable adoption.

Explainability – Gone are the days when the most respected solutions were those that stakeholders could not comprehend. Today, unless you can unravel the intricacies of your own solution, the solution is likely to find very few takers. Even if you are successful in convincing the first-line stakeholders, you might face trouble down the line when some of the results are not on expected lines and hence no one is able to figure out what has gone wrong.

Inclusivity – This is perhaps the most overlooked piece. AI solutions should ensure that all strata of the population are represented in the data being used.

Numerous examples in the recent past have shown that if any of these tenets are ignored, it might prove problematic for any AI-enabled product. Imagine a recommendation system for personalised beauty products. If the system persists in recommending non-intuitive solutions for a particular segment of the population, then—especially in today's age where negative publicity spreads very quickly—the product might become unpopular as people from that segment of the population take to public forums to express their displeasure. The same goes for an AI chatbot that does not take language sensitivities into consideration or is not strong on fairness and inclusivity towards its users. The chatbot will soon have to be discontinued.

Or imagine if an AI solution keeps allocating low scores to a startup headed by a member of a marginalised community who is looking for funding. Such a solution might cause venture capitalists to miss out on a wonderful investment opportunity on grounds that have nothing to do with business outcomes.

Lower adoption, discontinued products, and the rejection of good business proposals are only some of the outcomes of ignoring the tenets of responsible AI. These could easily be avoided if AI solutions pay attention to the principles of responsible AI – the need to cater to all segments of the population through the data being used, de-biasing the pre-conceived notions present in data, and checking to ensure that no form of bias is impacting predictions as well. Ensuring that these dimensions of responsible AI are front and center when developing any solution will ensure higher chances of adoption of the solution in a holistic and sustainable manner, and consequently win the confidence of end consumers as well.

Growth without guiding principles is seldom sustainable and the field of AI is no exception. Mainstream AI is still young, and numerous global challenges are screaming to be solved. The AI industry is too smart to ignore the risks of unreliable AI, and thus the frameworks and guidelines around responsible AI are being drafted at a rapid pace. These will go a long way towards securing the confidence of the mainstream consumer on AI-enabled products, thus helping drive exponential adoption.

Author:

 The author Devesh Raj is a Director with the AI Center of Excellence of the Analytics and Research group at Fidelity Investments India. The group works on the latest in fintech to deliver solutions to Fidelity's customers. Read more on the company's website.

Devesh is also a member of Algorithm Review Board at Fidelity Investments, reviewing data science solutions from multiple lenses, including AI ethics and explainability. He leads the India Data Science team of the Fidelity Institutional group. Devesh has also led AI/ML teams at Fortune 100 companies and built AI-enabled products for use across the US, EU, South East Asia, and Africa.

Views expressed are as of the date indicated and may change. Unless otherwise noted, the opinions provided are those of the author, and not necessarily those of Fidelity Investments.

Links to third-party web sites may be shared on this page. Those sites are unaffiliated with Fidelity. Fidelity has not been involved in the preparation of the content supplied at the unaffiliated site and does not guarantee or assume any responsibility for its content.