Artificial intelligence and machine learning continue to make a profound impact on the medical device and diagnostics industry. So what AI trends can medtech (along with many other industries) expect in 2020? Top AI experts at New York, NY-based Fractal Analytics, one of the largest AI providers in the world, shared the following predictions.
1. AI Will Be More Clearly Defined
"There have been numerous AI (or actually, ML) related use-cases in 2019 and lots of lessons learned, including that we are still at very early stages of ML usage, especially in business applications, and that we need to treat this phase as we did with many new methodologies that came before – identify the right problem, test a new approach, identify its applicability, iterate until clear advantage (vs. prior methods) is established," said Eugene Roytburg, managing partner of Fractal's Strategic Center.
Roytburg predicts that AI and ML will be more clearly defined as a part of broader analytics and will have better-defined application areas and value creation. "Many companies have been confused about the two, and if and how they are connected. While the companies pushed for many ML tests and use-cases mostly driven by 'artificial' pressure, the 'traditional' analytics development took a back seat," he said.
Next year, Roytburg said, lots of individual ML use-cases will give ground to better thought-out usages of "traditional" advanced analytics vs. ML.
"In 2020, most companies will come to their senses and consider AI 'craze' not as a singular development, but more connected to overall analytics strategy and transformation to optimize their existing analytics efforts while setting up the right infrastructure and governance to expand AL potential," he said.
2. A New Era of Human-Machine Interaction
Roytburg also predicted that we'll see more of a true human-machine interaction in which companies will leverage both objective data and analytics-driven insights, while at the same time incorporating human intuition and experience to learn from both worlds.
Similarly, Soudip Roy Chowdhary, CEO of Eugenie.ai at Fractal, said 2020 is going to see a new era of the human-machine confluence with further progress in the areas of deep learning especially in the reinforcement learning and generic adversarial network or remote process automation.
"Adding to it, the adoption of the AI techniques research community would continue to be invested in finding out new techniques by which they can explain the algorithmic decisions, transforming the black-box approach of AI to glass-box," Chowdhary said.
3. Design Will Still Be Important
Chowdhary also predicted that design will continue to play an important role in ensuring the wide adoption of intelligent systems to augment and extend human capabilities within organizations.
4. More Focus on Privacy-Aware AI Ecosystem
Privacy will continue to remain a concern for consumer-facing AI applications, Chowdhary said. He expects to see a substantial rise in research efforts related to building a privacy-aware AI ecosystem and enabling fairness in AI algorithms.
5. Demand for AI Will Still Outpace AI Talent
John LaRocca, managing director of Europe and North America operations at Fractal, said the demand for AI solutions will continue to outpace the availability of AI talent, and businesses will adapt by enabling more applications to be developed by non-AI professionals, resulting in the socialization of the process.
"Non-AI practitioners, such as knowledge workers and analysts, who are not skilled AI practitioners (but have great domain expertise), will start to develop rudimentary applications aided by automated AI engines," he said. "The onus will be on corporate training programs to retrain/upskill these new practitioners and on IT to enable them with automated AI environments that use AI itself (e.g., machine learning apps to help develops train models without having to write code)."
LaRocca also noted that this is not unlike the historical lifecycle of analytics, and it will similarly benefit everyone in the ecosystem because businesses will expand their capacity to develop and benefit from AI apps, AI experts will be working on truly leading-edge applications and newly upskilled non-AI practitioners will contribute more and have more marketable skills.
6. Better AI That Really Works, with Built-for-AI Hardware
Suraj Amonkar, a fellow who works with Fractal's AI @ Scale solution, predicted that the AI community will continue to make progress in building algorithms that train faster, train with lesser data, and generalize better. "The use of newer algorithms for data augmentation, few-shot/zero-shot learning will make the cumbersome, deep-learning training process easier and developments in feature representations and generative networks will work towards making models more generalizable," Amonkar said. "The use of complex/hybrid series of algorithms to achieve tasks will help build models that scale for complex 'real-world' scenarios. The use of self-supervised methods will hasten the progress of generic models. The availability of generic 'out-of-the-box' models in machine-vision and NLP will continue to evolve fast - but the need for building customized models for real-world challenges would remain. The use of multi-agent systems would evolve with the need to move towards more generic intelligence capabilities."
Amonkar also predicts that the use of both cloud-based and edge-based technologies will expand, and both these ecosystems will cater to use-cases that make sense for them. The use of micro-services frameworks, auto-scaling, and containerization will continue to deliver scalable AI ecosystems, and the use of edge devices will help bring real-time use-cases alive, Amonkar said. "Hybrid deployment models [will] be required in solving many of these complex deployments, and the use of in-memory and distributed storage/processing frameworks would continue to power AI systems increasingly," he said.
By the same token, Amonkar said there will be an increase in the development of special-purpose AI chips and hardware will allow tighter integration on the system. "The use of built-for-AI hardware would open vast possibilities to the amount of processing power that AI algorithms can leverage and would provide a major leap towards next-generation algorithms and systems."
7. Getting the Data Right First
Sandeep Dutta, chief practice officer for the Asia Pacific region at Fractal, said that 2019 saw a clear shift in enterprise resources toward getting the data strategy right before launching large AI projects. "Many companies focused on creating enterprise data lakes on the cloud that can help get good, reliable quality data sets in place," he said. "We are expecting to see this trend accentuate in 2020, with many centralized analytics CoEs first focusing on getting the data right."
He added that more and more analytics CoEs will be challenged to prove the ROI and impact of the solutions they create. "Organizations will look at incorporating design thinking to ensure that the user is in the middle and problems that are important for the business get prioritized, thereby improving the adoption of the AI and analytics solutions," he said. "The operationalization of the solution will be as important as solution development."
8. Deep Fake Technology Will Challenge All Industries
"Deep fake technology will continue to evolve very rapidly and will emerge as a big challenge for everything from the entertainment industry to politics," Dutta said. "Apps like 'Zao' from China, have already demonstrated the rapid pace at which deep fake is progressing, and this will pick up even more speed in 2020. The AI industry needs to fight back with better algorithms to detect deep fakes."
9. People-Centric AI Will be More Important
Parameswaran Venkataraman, the chief design officer at Fractal, said 2020 will bring more people-centric AI.
"Until now, the narrative about AI has largely been about how AI will replace human beings, In reality, several applications of AI are focused on augmenting and improving the way human beings work, rather than replacing their work. And therefore, focusing on the ‘people’ (not just ‘users’) who will use or be impacted directly / indirectly through the applications of AI, will become more important. This will mean focusing a lot more on understanding people, their current scenarios, behavior and needs, and how AI can help them in their goals. This will also mean designing AI solutions that are simple, intuitive, and delightful in their experience."