Machines Learning In The Wild


Following up on my post about the Bot Craze in machine learning and AI, I wanted to call out a few less obvious areas where increased activity in developing relevant machine learning can have a profound impact:

The Inbox

Google will surely continue to make strides here. We’ve surrounded so much of our information, personal and professional, to Gmail that improving sorting, logging, prioritization and archiving seems inevitable. Given that our inboxes are historical data treasure-troves, there is opportunity for startups to address this complex classification challenge right now.

Genetics & Predictive Diagnosis

While endlessly complex, many of the known cause and effect relationships between symptoms and illness/disease are ideal scenarios for machine learning. With large, existing data sample sets, predicting things like future medical issues or hospital re-admittance is possible. There’s further potential for a more sophisticated neural network where anomaly detection expands our understanding of root causes of disease and illness. There are numerous companies working on this problem, both large and small.

Legal Research

With a growing percentage of historical legal records available online and a trend in courts putting all of their trial information online, ML could save the army of associates that firms employ time and reduce opaque billable hours to clients by pinpointing relevant precedents for any case.

Credit Scores & Alternative Lending

Considering anomaly detection in a different light, a complex neural net should be capable of a more sophisticated and accurate profile of a loan applicant. This deeper and smarter analysis could be a profound lifeline for individuals without credit history or who have had any negative financial issues that in today often disqualify them outright.

Everyday Nutrition

A challenge to machine intelligence for human nutrition is input tracking. We have to be honest and diligent about recording what we consume. However, with a more efficient input, machine-powered nutritional recommendation based on our health, present and future, our dietary restrictions and any fitness or wellness goals could be a transparent knowledge base that we don’t have when deciding what to order at lunch or cook at home.

For many of these examples, machines would clearly be more powerful and efficient than the human brain once they’ve feasted upon the relevant data required. Access to these data stockpiles is not a given. For many sectors, words like scattered, offline and/or highly regulated describe the inhibitors to data access that is pre-requisite for machine learning. This is part of the reason why data network effects have become a hot topic. This status quo also encouraged entrepreneurs to build products with immediate, obvious value for end-users or customers that immediately access and compile data on the heels of delivering upfront value (Ex: 23&Me).

Another challenge to the examples above is the clarity and consistency of data organization. Messy, unpredictable data can be a nightmare to index and process. Depending on the volume of data and types if input sources, entire interstitial systems need to be placed in front of any new product attempting analysis. Even when captured, raw data is tough to translate into uniform sets.

There are a number of companies — startups and corporations — working on bringing machine learning to the sectors outlined above. My intent in describing the specific, potential value in these areas is to express that conversational bots aren’t the sum total of A.I. More likely, they are a major step in increasing mainstream awareness of and comfort with intelligent machines. While for now less marketable than bots, the above ML applications can bring an accuracy and efficiency to industries that is far beyond the capability of the human mind.

Tim DevaneComment