Artificial intelligence is one of the fastest moving and able industries. Think of all the things that couldn't be done a few years ago: deepfakes, AI-powered machine translation, and bots that can learn very complex games.
Testing the possibilities to predict the future of AI will never hurt. We asked scientists and AI thinkers what they think will happen in the AI space in the coming year. Here's what you need to know.
AI makes health care more accurate and less expensive
Philips' Chief Innovation & Strategy Officer, Geron Tass, told TNW: “The core impact of AI in 2020 is transforming health care workflows for the benefit of patients and health care professionals, reducing costs. Its ability to interpret and analyze data in real-time data from multiple hospital data streams, electronic health records, emergency department access, equipment usage, staffing levels, etc., helps to increase efficiency and care. "
It comes in the form of automated experimentation of optimized scheduling, automated reporting and equipment settings that “optimize the functionality of an individual clinic and the patient's condition - characteristics of the patient and staff experience, resulting in better outcomes and reduced costs. "
There is a lot of waste in many health care systems associated with complex administrative processes, lack of preventive care and high diagnoses and treatment. These are areas where AI can really make a difference, ”Tass told TNW. "Additionally, one of the most promising applications for AI is in the Command Centers area that optimizes patient flow and resource allocation."
Philips is a key player in developing existing AI-enabled applications. Currently, every two Phillips researchers around the world are working with data science and AI, launching new ways to apply this revolutionary technology in healthcare.
For example, Taz explained how integrating AI with expert clinical and domain knowledge can accelerate simple and simple yes / no diagnoses - not just overhauling clinics, but also spending more time on difficult and often complex decisions around them. Personal Patient Care: "AI-enabled systems track, predict, and support patient severity and availability of medical personnel, ICU beds, operating rooms, diagnostic and therapeutic equipment."
Further attention may be drawn to interpretation and belief
"2020 will be the year of AI credibility," Karthik Ramakrishnan, head of AI and AI at Element AI, told TNW. The earliest principles of AI and risk management emerged in 2019, and there have been previous attempts to apply these principles in toolkits and other research approaches. The concept of explanation (which can explain the forces behind AI-based decisions) is increasingly popular. "
In 2019, AI will definitely focus on ethics. Earlier this year, the European Commission published seven guidelines for the development of ethical AI. Element AI, founded by Joshua Benzio, one of the founders of the in-depth study, has pioneered the creation of data trusts and the ethical use of AI in collaboration with the Mozilla Foundation. Big tech companies like Microsoft and Google have also taken steps to change their AI development to meet ethical standards.
Ramakrishnan reminds us of the growing interest in ethical AI after some failures around credibility and AI in the market, such as the recent Apple Pay roll-out or the recent Cambridge Analytica scandal.
“In 2020, organizations will focus more on AI trust, whether they are ready or not. We hope that Visis will pay attention as new startups are emerging to support solutions, ”said Ramakrishnan.
AI becomes less data-hungry
“We are seeing an increase in data synthesis methods to address data challenges in AI,” Rana El Kaliobi, CEO and co-founder of Affiteva, told TNW. Deep learning methods are data-hungry, which means that AI algorithms built on deep learning can only work when trained and validate large amounts of data. Companies that develop AI can have the right kind of data and access to the data they need.
"Many researchers in the AI space have begun to try and use emerging data synthesis techniques to overcome the limitations of real-world data available to them. With these techniques, companies can take existing data and integrate it to create new data," said El Kaliobi.
Take the automotive industry as an example. As the industry works to develop sophisticated driver safety features and personalize the transportation experience, there is much interest in understanding what is happening to people inside the vehicle. However, collecting real-world driver data is difficult, expensive, and time-consuming. Helps address data synthesis - for example, if you have a video that you can drive in my car, you can use that data to create new data, such as to tilt my head or wear a hat or sunglasses, ”added El Kalyubi.
Thanks to advances in fields such as Generative Adversary Networks (GAN), many areas of AI research can now integrate their own training data. However, data synthesis does not eliminate the need to collect real-world data, El Kaliobi recalls: “[Real data] is always key to developing accurate AI algorithms. However can improve those data sets. "
Improved accuracy and efficiency of neural networks
"Neural network architectures grow in size and depth, deliver more accurate results, and perform better at simulating human performance in tasks involving data analysis," Kate Senko, associate professor in the Department of Computer Science at Boston University, told TNW. "At the same time, the methods of improving the efficiency of neural networks are also improving and we are seeing more real-time, energy-efficient networks running on smaller devices."
Senko predicts that neural generation techniques such as Deep Fake will enhance and create more realistic manipulations of text, photos, videos, audio and other multimedia that humans cannot find. The creation and discovery of deepfakes has already haunted cat and mouse.
As AI gets into more and more areas, new problems and concerns emerge. "As these AI systems become more widely used in the community, they are more closely scrutinized for the credibility and bias behind them. For example, more local governments are considering banning surveillance from AI due to privacy and affordability," Jainko said.
Senko is Director of BU's Computer Vision and Learning Group and has a long history of researching visual AI algorithms. In 2018, she helped develop RISE, a method that closely monitors decisions made by computer vision algorithms.
Automated AI development
"In 2020, IBM expects to see significant innovations in what's called 'AI for AI': AI will be used to help automate the steps and processes involved in the life cycle of creating, implementing, and maintaining AI models. Add AI to the enterprise," says IBM Research AI VP of Sriram Raghavan said.
AI automating has become a growing research and development field over the past few years. Google's AutoML simplifies the process of creating machine learning models and makes the technology available to a wider audience. Earlier this year, IBM launched AutoAI, a platform that automates data creation, model development, feature engineering, and hyperparameter optimization.
"In addition, we will begin to see more examples of neurosymbolic AI use, which can use statistical data-driven approaches to combine robust knowledge representation and rational methods with fewer data to provide more detailed and rigorous AI," Raghavan told TNW.
One example is the neurosymbolic concept learner, a hybrid AI model developed by IBM and MIT researchers. NSCL combines classical rule-based AI and neural networks to solve some local problems of existing AI models, including large AI requirements and lack of explanation.
AI in construction
"2020 will be the year that the manufacturing industry adopts AI to modernize the manufacturing sector," said Maximiliano Versace, CEO, and co-founder of Neura. The biggest challenge for the manufacturing industry is quality control. Product managers struggle to inspect each individual product or part, while bulk orders expire. "
By integrating the workflow solutions as part of the AI, AI adds to the challenge and solve: versus believes: "We have a policy similar to that power screwdriver drill, AI construction industry to improve existing processes, reduces the risk of heavy and dangerous jobs, and leads the industry with innovative production workers in Middle East Free up time to focus on growth. "
“Manufacturers are going to the brink,” Versace adds. As AI and data become centralized, manufacturers will have to pay huge fees to top cloud providers to access the systems they operate and run. The challenges of cloud-based AI have led to innovations in building edge AI, software, and hardware that can implement AI algorithms without the need for a link to the cloud.
New ways to practice AI can be implemented and modified. As we enter the New Year, more and more manufacturers are turning the corner to create data, reduce latency issues, and reduce massive cloud fees. By deploying AI wherever necessary (at the end), manufacturers can retain ownership of their data, ”Versace told TNW.
The geopolitical implications of AI
“AI will continue to be a national military and financial security issue in 2020 and beyond,” said Ishan Manaktala, CEO of Symphony AIASDI. Governments are investing heavily in AI to become the next competitive front. China has invested more than $ 140 billion, while the UK, France and the rest of Europe have invested more than $ 25 billion in AI programs. The U.S., which is a late start, will spend about $ 2 billion for AI in 2019 and more than $ 4 billion in 2020.
Manaktala added, “But experts are calling for greater investment and warned that the US is still lagging behind. According to the National Security Commission on Artificial Intelligence, China is expected to outperform US R&D spending over the next decade. In its preliminary report, the NSCAI outlined five aspects of investing in ARCA & D, applying AI to national security operations, training and recruiting AI talent, protecting US technical successes, and Marshall's global integration. "
AI in Drug Discovery
Automated visual processes will automatically improve drug detection in 2020, as visual AI can better monitor and detect cellular drug interactions, ”Emra Gultekin, CEO at Chuch, told TNW. Currently, clinical trials have been wasted for years, with researchers taking notes, putting them in spreadsheets and submitting them to the FDA for approval. Instead, more accurate analysis of AI can lead to drastically faster drug discovery. "
ష Development takes up to 12 years and involves the collective effort of thousands of researchers. The development of new drugs has crossed the 1 billion mark. AI algorithms are expected to accelerate the process of testing and data collection in drug discovery.
“In addition, cell counting is a major issue in biological research, not cell innovation. People go through a microscope or count cells in the hands of those who click in front of screens. There are expensive machines that try to be precise. But visual AI platforms can do this with 99% accuracy, ”Gultekin said.