AI That Really Works – Part 2: Ripe Sectors & Functions

Every industry sector and business function
will be upended by AI, it’s just a matter of when and how fast

This article addresses the second part of the question of where AI reality works. There has been a lot of attention in the media about how some industries are being disrupted by AI. It’s true to say that almost every sector of the economy will be affected eventually. The business question is to know which sectors and functions are ripe for adoption of AI sooner, and how fundamentally and fast AI may change things.

Bain & Company puts industry sectors into three buckets in terms of AI disruption: Optimization, Transformation and Reinvention.

In terms of optimization, almost every industry sector or business function can benefit from the use of AI. Whether in cost reduction or productivity gains, learning systems will be able to improve the efficiency of most business functions. Optimization is simply doing what they do now, but just faster or cheaper. This is essentially AI at the specific task level. For example, agriculture, manufacturing, and supply chain logistics have lots of potential to be optimized. AI will allow for new levels of automation, including having robots delicately pick only the ripe strawberries without damaging the plant for example.

The second level of disruption is transformation. This is using the superhuman qualities of AI to do things that were not possible before. Typically it means using AI in tandem with people, working closely together. Functions like sales, customer service and relationship marketing augmented with AI make for a powerful combination. Healthcare is a good example of a sector undergoing an AI fueled transformation, from providers to pharma companies. Doctors will be working side by side with AI systems to diagnose and treat diseases. Researchers will use tools not imagined before to sift through millions of potential drug candidates. Likewise in the finance sector, fraud detection and cybersecurity is becoming a war of bots, defenders struggling to outsmart attackers in a constant battle.

The third level of disruption, Reinvention, means that the basic structure and economics of the sector is fundamentally changed. This means sometimes turning over whole functions to an AI system. Transportation, with AI-powered self-driving cars, will likely undergo substantial changes in capital requirements and business models. Likewise, retail and consumer marketing may become unrecognizable in a very short amount of time. These are ambitious claims, and the technology is not fully there yet, but the potential to make radical change in these sectors is very real.

Many factors affect this assessment, so there are no clear cut rules. Some sectors/ functions have better data availability, especially quantity, quality and labeling necessary for machine learning. Also, if the risk of failure is more acceptable, or at least can be mitigated with other efforts, may make AI applications more amenable. Tolerance for some types of errors and probabilities. On the economics, the value of the AI application (e.g., avoided costs or increased revenue). Organizational readiness to adopt technology, or data science driven decision making. Finally there are some areas in which ethical/ policy concerns are difficult to navigate.

Agriculture is being optimized through automated precise harvesting,closed loop soil sensors/ fertilization and predictive analytics to dramatically improve productivity

One business sector being optimized by AI is agriculture. There are scores of startups building AI solutions in this sector. Automation of harvesting is nothing new with crops like corn or wheat, but AI is needed to harvest more delicate crops. Unlike mechanical corn harvesters, strawberry picking robots like the one developed by HarvestCroo uses computer vision and advanced robotics to identify and pick only the ripe berries without damaging the plant. This is back-breaking and delicate work that up until recently could only be done by humans.

In addition, using soil sensor data and images from drones and satellites, AI can monitor crop and soil conditions. In this way, pesticides and fertilizers can be deployed much more precisely with GPS driven robot tractors. This saves a lot of money, increases yields and makes a healthier environment. Finally, AI predictive analytics can be used to manage resources and deal with risk, everything from the weather to commodity markets.

Likewise manufacturing and supply chains are being optimized by AI to reduce waste, defects, inventories and logistic costs

Another sector being optimized by AI is manufacturing. Business functions like supply chain and logistics are good examples of AI optimization. We already have discussed how warehouse operations are being improved using robots by companies like Amazon.

In addition, co-bots will broaden the ability for AI-driven automation into less structured manufacturing settings. Finally, the process of quality control is being augmented by AI. Using computer vision and anomaly detection techniques, AI is able to check for flaws in manufactured products.

Many AI startups in this sector are providing real value to their business customers through a combined machine learning product and consulting service. This makes for an attractive business model for the AI startup by providing immediate revenue, as well as yields better results for the customer. Overtime, the startup can generalize the product offering to get more leverage and reduce human involvement in the deployment.

One example of this approach is Landing AI which offers a customizable AI platform to its business customers to address several categories of problems including defect detection. They are providing AI as a software as a service (or SaaS) as well as a professional consulting services to adapt their system to the customers’ needs.

Healthcare is being transformed by AI, from drug discovery to disease diagnosis. But expect medical professionals to stay in the loop for the foreseeable future.

One sector going through an AI-driven transformation is Healthcare. There are hundreds of AI startups in this space. Because of what is at stake, AI is unlikely to completely replace human judgement in healthcare anytime soon. Having a medical professional in the middle to validate the AI recommendations will make a potent partnership, leading to better health outcomes, lower costs and more rewarding work.

Diagnosis is one of the most successful applications for AI in healthcare. AI can be used to ensure clinically proven protocols are complied with, or if a different action is taken, to help capture the reasoning. The startup HealthTensor which I mentioned during the intro, is using AI to do just that. Finally, AI is being utilized to supercharge the process of drug discovery.

Accurate and early diagnosis is an essential part of healthcare. AI can be a powerful ally in the process of medical diagnosis. In particular, many startups are using computer vision technology to review medical images. This can be x-rays, MRIs, biopsy slides or even photos taken by the patient using a smartphone. These are well-suited to be classification problems where we use data to train the model and predict the probability that there is or is not a disease present.

One startup, Paige.ai is combining biopsy slide image data with clinical notes to improve this prediction.

Other companies use low-quality imaging to identify early signs of disease, in this case using a smartphone by patients to detect melanoma. Using deep learning convolutional neural networks, even the relatively low quality images yielded results better than experienced dermatologists. Such applications of AI make sense depending on the cost-benefit of a positive finding relative to a false positive. In the case of melanoma, early detection is critical and the follow-on cost and risk is very low.

One example of the third area of healthcare transformation from AI is drug discovery and development. The startup company Atomwise is using a series of AI techniques to help reduce the number of targets, and thus greatly reduce the cost of discovering new drug therapies. Atomwise is using AI to look at the shape of molecules to infer the likely function compared to similarly shaped molecules. In doing so, they claim to improve the hit rate by up to 10,000 times and deliver accuracy comparable to wet lab experiments.

Finance is also being transformed by AI, from pattern recognizing fraud detection and credit assessment, to computer trading and investment management.

Finance is another sector being completely transformed by AI. It’s one of the most crowded spaces for startups, supercharging everything from banking, investing, credit and insurance services. Here are just a few areas of successful adoption of AI in the finance area.

The first and foremost successful AI application in this area is fraud detection,

AI applied to insurance and credit, including risk assessment and management, has also proven to be useful, and finally Market analysis and investment management, including algorithmic trading, are areas ripe for AI.

Fraud detection is a natural AI application. Many companies are already using AI to detect fraudulent transactions on their platforms, and this is expected to almost double in the next couple of years. The best approaches use both supervised and unsupervised machine learning models. First, supervised learning is trained on historically known fraud attempts. Second, the supervised model is combined with unsupervised models to cluster and detect emerging approaches. In both cases, companies are tuning their models based on a wide set of data, not just the usual indicators for fraud.

Auto and transportation sector may be completely reinvented by autonomous vehicles, utterly changing the economics and value add for the industry to the point of being unrecognizable within a decade

There are some sectors that are already showing signs of being completely reinvented by AI. Transportation is one of those sectors. The potential of fully autonomous vehicles will have a profound impact on the transportation sector, completely changing the economics of the related businesses. Big tech and auto manufacturing companies are investing billions into the space, supported by a swarm of startups providing technology and services. However, the underlying technology has a long way to go, but there are promising off ramps along the way.

Companies like Google’s Waymo are shooting for the stars, working to build completely autonomous vehicles. The service is already being tested on roads in Phoenix and parts of California. Waymo is not driving without a safety driver even though they legally could. Nonetheless, their self-driving cars drove autonomously for more than 11 thousand miles before a human needed to take over. This is approaching what is referred to as level five autonomy, which means that no human interaction should be necessary to go from point A to point B under any driving conditions. Even level 4 seems to be a stretch though, which would safely stop the car if weather, construction or other conditions interfered. But either level 4 or 5 seems a long way off unless cities adopt designated areas or times only for self-driving cars, and they are limited to low speeds.

Meanwhile, car companies like Tesla are working on assisted driving (referred to as level 2 or 3 autonomy) which requires intermittent involvement by humans. The advantage is that these companies are accumulating huge amounts of data about roads and driving conditions. Tesla is using relatively low-cost sensors, relying on cameras and radar only.

The biggest challenge with this assisted driving approach is that it’s hard to get humans to stay available to re-engage when needed. Humans become complacent and distracted when not engaged in an activity. Google had this problem when they initially started testing their semi-autonomous cars with employees. In a short amount of time, test drivers started texting, applying makeup, using their laptop and even falling asleep, causing the company to shut down the test and instead shoot for complete autonomy. Tesla and other autopilots will need to develop a workable interface that keeps humans ready to re-engage, or otherwise bring the vehicle quickly to a safe stop.

The reality is that there are too many corner cases in driving to expect complete autonomy. These corner cases include severe weather conditions, like heavy fog or rain, road construction or accidents. In these situations, a human will need to reacquire control very quickly and seamlessly.

One solution for this has been Starsky Robotics, a startup that allows remote controls of trucks (UPDATE: Starsky filed for bankruptcy in March 2020). Using this approach, the AI could do most of the highway driving so long as conditions are within operating parameters. When it is time for the truck to exit the highway or driving conditions become a problem, then the remote driver can take control of the truck, and still gets to go home to dinner with his or her family.

Retail and consumer marketing will be a completely different sector within 10 years, in part due to AI.

Finally, the retail sector is also going through the beginnings of an AI-powered reinvention. This means retailers may have a completely different value proposition and economics for their business. As I mentioned during the intro, companies like Adept mind are using AI to help people find products while leveraging the knowledge of retail staff. AI-powered computer vision is being used by Amazon Go, an experimental convenience store that allows people to shop in stores without needing to checkout.

These are just some examples of sectors going through profound changes due to the adoption of AI, but we are still in the early days.