[PULSE] Why we need diversity in AI development

[PULSE] Why we need diversity in AI development
Trust is a universal concept and like love, can be fickle, fleeting and fuzzy. Some trust is implicit – when I board a bus, I trust the driver to drive safely. I know nothing about the driver or the bus, I could be in a foreign country, but I trust. I am shocked if they prove me wrong.

Some trust takes work. I have learned to trust a certain hairdresser, a particular sandwich maker, my husband. This took time and effort from both parties – repeated execution towards a set goal and consistently meeting them.

Aneeza HaleemGuest Author

And of course, there are situations that require contextual trust. I trust my sandwich maker to give me a great sandwich, but can I trust him to do my eyebrows? I trust my husband, but not necessarily with cheesecake.

It is these fuzzy situations that tend to trip up even the smartest artificial intelligence (AI) workforce. We have established tried-and-true AI workflows – back-office operations, analytics and advanced computing. We have successfully tackled using AI in newer areas, such as tiered contextual responses, voice recognition, biometrics and natural language processing.

The fuzziness increases in emerging areas of AI use, including one where it’s especially common in mortgage banking – customer engagement using sentiment analysis and advanced contextual cues.

AI works best in black and white

It’s been repeatedly proven over the last few decades that if you focus an AI on a definitive task – chess, calculus and soil management – it excels. But contextual knowledge remains an enigma, mainly because of our limitations. The code we write and the logic we create is influenced by the bubbles we live in.

If you have a Google news feed, or a social media account, you have experienced AI working on a definitive task. A news and media feeds that provides content with similar attributes keeps you engaged.

As New York Times Tech Columnist Kevin Roose uncovers in his chilling Rabbit Hole podcast, what it does not do is provide you with diversity of thought.

The AI driving those feeds does not show you news and media feeds that might expand your view. Nothing you see is going to challenge your beliefs. That is a critical point because our experiences and beliefs – both firsthand and secondhand through books, friends, movies, and feeds – all play a significant role in the knowledge we hold and impart.

But the best AI is in color

The idea that diversity is needed in any AI development to ensure a roundness, a level of acceptable morality and humanity may not be new, but it continues to be an issue.

Every iteration from Nikon’s failure of recognizing Asian faces to Microsoft’s disastrous troll-taught bot Tay, has widened our perspective. AI needs interaction with people, but individuals bring positive and negative experiences and beliefs to those interactions.

How do we capture the many cultural social norms of the American melting pot? We know to use larger datasets, look for fringe cases, diversify our focus groups and refine through larger and larger testing.

In mortgage banking, AI does its best work answering questions with exact answers for questions such as, “What’s the most I can borrow if I use Fannie Mae’s HomeReady?”

Where AI still falls short is in areas that our mortgage loan originators excel, the fuzzy areas where beliefs, experience and social norms influence our financial decisions. AI can offer pros and cons to answer the question, “Is it cheaper to use Fannie Mae’s HomeReady or my state bond program, or both?”

What it cannot advise on is how your relationship may change after using your father-in-law’s income to qualify for HomeReady.

AI gets faster, smarter, better every day

Ray Kurzweil waxes eloquently about human history’s law of accelerating returns – advanced societies have the ability to progress faster since they are more advanced. Marty McFly was shocked when he went back 30 years from 1985 to 1955 in the movie “Back to the Future.” But if he were to go back 30 years today – from 2020 to 1990 – his mind would be blown! Smart phones, internet, social media, mumble rap, K Pop – the pace of change would be even more evident.

History dictates that we are fast on our way to a truly intelligent bot. There are many schools of thought on how we will reach this goal – evolutionary algorithms where we mimic natural selection by combining logic that we “deem” accurate. Maybe self-learning algorithms that enable the AI to code and improve its own architecture. Whatever the path, the goal is in sight.

Where bot integration is easiest right now

While we await the intelligent bot, perhaps the easiest integration for AI in mortgage banking is into our workforce. The value proposition of using AI to increase efficiency and optimize the current workforce has been established.

Many lenders who tout accelerated close times leverage varying degrees of AI to supplement their teams. Any activity that may be defined as business rules (no matter how nested) can be seamlessly executed. The most common tried-and-tested AI assisted workflows now include ordering disclosures, integrations with third-party services, AUS/DU, and QA/QC to verify values across systems and documents.

Newer workflows like intelligent NLP chatbots are getting smarter and better able to provide contextual advice. The scenarios they respond to still need to be defined by a human but the bot is able to retain context through multiple levels and understand slang, emojis, and some levels of content.

The difference of providing not just an animated version of a FAQ, but an actual advisor who can walk a customer through multiple scenarios is incredible. The customer experiences  hyper-personalized, focused attention delivered on their schedule. The  gains a huge lift in productivity from having a trained “team member” who works 24/7 for a fraction of the cost.

Careful consideration needs to be made on how the bot is integrated. Focused tasks with clear business rules are easy solutions and the industry is filled with potential providers.

Contextual situation requires more thought and grooming – how does the bot fit your representation of your company? It is also important that the bot align with your corporate values and mission. You should pay attention to how it reacts with not only your core demographic but every customer and potential prospect. Also, companies need to determine how to manage interactions and how bots will update as customer beliefs, views and social norms change.

With where bots are right now, we are really looking for trust in our teams to design, nurture, and tailor a bot to have the best interests of our companies, our customers, and our business partners in mind. By focusing on the fuzzy, we can create a contextual experience that builds the trust we want our end-users to have in us.

The views and opinions expressed in this article are those of the author and do not necessarily reflect or represent the views, policy, or position of Planet Home Lending, LLC.
The post [PULSE] Why we need diversity in AI development appeared first on HousingWire.
Source: https://www.housingwire.com/rss