During my flight home from a recent meeting, I was reflecting on the feedback we received from a client with a strong internal data science team, and my mind drifted to the impact Yahoo Finance had on the stock market over 20 years ago.
Understanding why a client meeting about our AI-powered text analytics triggered memories of Yahoo Finance requires some explanation.
Yes, artificial intelligence is changing how we do things …
AI-powered algorithms increasingly influence our daily lives. Recommended songs prompted by Spotify, facial recognition in security cameras – and the inexplicable fluctuations in my FICO score – are all powered by AI algorithms, developed by sophisticated data science teams and machine learning specialists.
Development of AI-powered algorithms is largely confined to highly-trained specialists who gather the millions of data points required to tag, train, and tweak models until the optimal solution is found – be it the song Spotify thinks you want to hear or the risk that I won’t pay my credit card bill. Whatever the question, given enough labeled data and enough time, humans have gotten pretty good at finding paths to optimized solutions. These algorithms are a huge leap forward for data science. Everything from agriculture to marketing to gaming is being transformed by AI-derived solutions.
… but AI currently has some significant weaknesses.
It’s true that AI can help solve problems too complex for humans. But many problems don’t have millions of data points to create a model for AI in the first place. Finding enough data to tag, train, and tweak for domain-specific issues is often not feasible.
Second, the tagging, training and tweaking process depends on a team of highly-valued data science and machine learning specialists. The scarcity of these specialists creates organizational bottlenecks, forcing prioritization of only the most critical business problems.
And finally, business users – those closest to the underlying problems everyone’s trying to solve – wait months to get a static solution based on stale data. Not to mention, any required updates must also be conducted by the data science team, meaning the involved and time-consuming process of incorporating new data, retagging, retraining, and readjusting the model.
The path forward? Democratize AI.
Organizations’ ability to use the full power of their data depends on debottlenecking data science teams. Mikael Paani, data specialist at Supercell, makes this point explicitly:
“From the get-go, we took the approach that everyone who can benefit from data should have access to data, and should have training in order to use that data for their work. So, instead of having a handful of really well-trained analysts, it’s a lot better to have 150 people who know how to use data to do their work.”
We’ve received similar feedback from other leading brands. Last year a financial services client told us it wanted to train dozens of its digital product managers on how to use Luminoso, allowing them to debottleneck their data science team and react quickly to customer feedback. Companies with digital products and services, which see changes in customer feedback with every software update and new feature, simply cannot depend on rigid models that take months to develop on data that is stale by the time it’s ready to use.
Organizations come to Luminoso because our technology, which is pretrained on an enormous natural language background space, bypasses the need to tag and train millions of data points. This means business-critical insights are available in a few minutes with only a few hundred data points. We’ve always been justifiably proud of the sophistication of our text analytics – and we still are. But feedback from our clients has pushed us to likewise focus on our ease-of-use so that everyone in an organization, not just the data science team, can use our software to instantly identify and solve emerging issues in unstructured text data.
Motivated by these observations, over the last year, we invested heavily in making our Luminoso Daylight® application much easier to use for the non-technical business user. We’ve built intuitive data analyses of easily-imported unstructured text directly into the user interface. And while we continue to advance the science in our product, we are just as proud to be a leader in the democratization of AI by putting the power of our software into the hands of the business user.
After we demoed our recent product improvements to Mikael Paani at Supercell, I asked which was more crucial: prioritizing advancements in the power of our technology or the ease of its use. He unequivocally emphasized that ease-of-use – and corresponding training materials for business users – is his priority at this time. Fortunately, he doesn’t have to choose: we’re advancing both our science and our accessibility in parallel.
So what does this have to do with Yahoo Finance?
In the 1990s, Yahoo Finance (and others) enabled anyone with a personal computer and an AOL account to quickly perform analyses and comparisons of public companies. Until that point, these analyses could only be done by companies with expensive Bloomberg terminals or similar analytical tools. Democratizing financial analytics gave rise to an informed individual investor who has since permanently changed global financial markets and opened the floodgates for an entire set of personal analytics tools.
As AI-powered solutions become more prominent, there’s fear that jobs and tasks will be eliminated. But let’s return to the Yahoo Finance example: providing financial analytics to individuals did not diminish the value of investment banks or institutional investors – it meant just the opposite. Freed from routine analytics, “quants”, hired by banks and hedge funds, focused on developing increasingly sophisticated trading algorithms that are not only beyond impactful in today’s markets, but also the subject of many popular books and movies.
Companies like Luminoso are similarly democratizing AI for our clients. By developing AI-powered analytical applications for business users, and eventually by individuals, AI will overcome its current bottlenecks and limitations and be able to solve problems for anyone with a laptop and WiFi connection.