Top 10 AI(Artificial Intelligence) trends of 2019
Most of us are wondering about “what’s next for AI in 2019 leading up to 2020?” Some of us might be wondering about “Singularity”. Others are still thinking, “AI is all hype and no action” (Although this subset is on the decline). All valid questions to ponder upon. Artificial Intelligence has made substantial progress in 2018. If you don’t believe me, look around yourself and notice all sorts of subtle uses of AI. The Gmail email response prompter, Alexa in your kitchen, the transcription of your voicemails, automatic voice calls to your hairdresser for your next appointment and of course self-driving cars on the roads of Utah! Apart from the above mentioned consumer use cases, industries saw the rise of applied AI. Take the case of Lyft. Lyft uses a series of deep models for the fraud detection that can be in the form of using stolen credit cards to generating transient rise requests. Meanwhile, Walmart uses AI models to increase operational efficiencies like sales tax refunds and audit processes. In higher education, universities like Penn State uses AI to help students find the shortest path to graduation. In a recent survey by Deloitte (Q3 2018), it was very clear that “Early adopters are ramping up their AI investments, launching more initiatives, and getting positive returns.” Here is a look at how AI is helping organizations. This is further accelerated by investments in AI startups that rose by 21% and will continue further in 2019. According to a joint report by PWC and CBInsights, funding for US based AI companies was close to $2.3 Billion. The graph below shows how the increase in AI funding over the last two years: Suffice to say that there would be an upward trajectory in 2019 as well. Let us look at the top 10 trends of AI in 2019: 1. Reinforcement learning: Reinforcement learning is different from supervised and unsupervised learning (the most common forms of machine learning). Supervised learning involves learning with labeled datasets to produce an output that is generic to that dataset (for example finding the price of a new house given the housing prices of a specific location). Unsupervised learning involves finding the connections between unlabeled data or clustering that data (think of a bunch of images that are not labeled but have parameters like color, size etc. and the program will return a result of whether the image is a fruit or an animal). Reinforcement learning is unlike the above methods; it is a framework that does not use the data recognition techniques mentioned above. Instead it uses experience-driven sequential decision-making. This method interacts with the environment to learn and move towards a goal that rewards the actions taken. Most game playing algorithms use reinforcement learning – determining the moves the computer should make to win the game. Without the need to specify all the rules of the game, the algorithm learning from playing the game repeatedly and exploring all possible options. The use of reinforcement learning is currently very limited (AlphaGo and some robots use it) but many industries are exploring the uses and would continue experimenting with it in 2019. Some of the industrial use cases that are under consideration are: · Higher education – use of reinforcement learning for personalized teaching and learning systems. · Healthcare – Determining treatment policies for chronic illnesses like schizophrenia, diabetes etc. · Finance – Although many models are currently under study, a live use case is JPMorgan Chase RL program called “LOXM” that is used for executing trades in the stock market. It is known to have increased the speed of the client orders that are executed at the best possible price. 2. Ethics in AI: This is a huge topic and a personal quest for me. “With great power comes great responsibility” This adage holds true to Artificial Intelligence applications. While there are debates and discussions about AI replacing human jobs, there is a fundamental question that needs to be answered. Are the autonomous and intelligent systems being designed in a manner that includes individual, community and society’s ethical values? 2018 saw many presentations on this topic. Watch this and this. This topic is only going to increase in 2019. What are some of the things being done to ensure ethical design of AI applications? According to the IEEE global initiative on ethics of autonomous and intelligent systems, we would need to have policy or standards around the following: · Legal accountability – Whether it is property laws or legal responsibility for harm caused by the AI systems. · Transparency – This is not only to reinforce the policy of data usage but also or reasonable access to the rules embedded in these systems along with audit trails. · Policies – Policies surrounding the impact and implications of these systems should be in place. · Embedding values into AI applications - This can be done by expressing norms in terms of obligations and prohibitions that can be expressed computationally. · Governance frameworks - Standards, processes and procedures in place to not infringe upon basic human rights. 3. Quantum Computing: On the onset, let me dispel any myths that we would make quantum computing incrementally better in 2019. Instead we would just be pushing a little in the right direction towards building better quantum computing devices. Although it would be small increment, it would still be a huge focal point in the area of AI. Quantum computers use quantum physics to compute calculations faster than any supercomputer today. We are well aware of how computers use bits and bytes. However unlike a regular computer, quantum computers use qubits (Quantum bits) to store information. We have a long way to go in terms of dealing with the challenges of quantum computing like maintaining coherence of the qubits or removing the unnecessary and noisy computations. 2019 would see more research on quantum computers and how to create strategies to reduce the error rates to make meaningful computations possible. According to Andrew Childs, Co-director, Joint Center for Quantum Information and Computer Science (QuICS), “Current error rates significantly limit the lengths of computations that can be performed, We’ll have to do a lot better if we want to do something interesting.” The interesting problems could be to solve almost unsolvable problems like climate change, presence of earth-like planets in the galaxy or our body’s ability to destroy cancer. 4. Convergence of AI and other emerging technologies: 2019 would see more examples of the convergence of AI with IoT and AI with Blockchain. In fact self-driving cars is not a practical possibility without IoT working closely with AI. The sensors used by a car to collect real-time data is enabled by the Internet of Things (IoT) and the programs used for decision-making is powered by AI models. Deep learning AI algorithms then take action as well as make decisions using this data. Some of them include path planning, eye-tracking to improve driver monitoring, natural language processing to understand voice commands and maybe even self-direct itself to a gas station when running low on fuel. The other powerful feature that these autonomous vehicles would have is the ability to communicate with each other so that traffic as a whole is optimized. Another integration of disruptive technologies is Blockchain and AI. We are well aware that Blockchain has challenges such as security and scalability and AI suffers from privacy and trust issues; these two can be combined to address these issues. Blockchain can power decentralized data marketplaces and help AI algorithms to be more transparent and trustworthy. There are several startups paving this way to democratize the AI training datasets. For example, Enigma is a startup that allows organizations to a secure data marketplace that users can subscribe to and consume via smart contracts. 5. Facial recognition: Whether it is Google winning the recent lawsuit or China’s SenseTime, Facial recognition has received a lot of negative press recently. However this technology would continue to grow in 2019. Facial recognition is a form of artificial intelligence application that helps in identifying a person using their digital image or patterns of their facial features. 2019 would see an increase in the usage of this technology with higher accuracy and reliability. We are already aware of Facebook’s Deepface program that is used to easily tag your friends and family in your photos. The popular iPhoneX is already using facial recognition as a digital password. With the boom in personalizing everything – from your shopping experience to advertising, this technology is going to be used more and more for biometric identification. This will continue to rise due to the non-invasive identification and the easy of deployment. Other use cases like payment processing through security checks as well as for law enforcement (in early detection and prevention of crime) would be on the rise. These next-generation image recognition technologies can be used for healthcare purposes as well – to follow through clinical trials as well as medical diagnostic procedures. Openwater, one the forerunners in imaging technologies, is pushing the boundaries of future devices that could read images from our brains! 6. Biased data: This topic is becoming increasingly important as machine learning models are being used for decision-making such as hiring, mortgage loans, prisoners released from parole or the type of social service benefits. Amazon is reported to have scrapped an internal hiring tool that increased bias against hiring women. Some of these biases are conscious while some are unconscious due to data that is used for training. For example, consider a fictitious case of the decision of promoting a woman. Historical data on employment may show women getting less promoted than men and hence create discriminatory AI applications. Many more examples have led to an increased emphasis of dealing with biased data in AI applications. As the usage of AI applications increases in 2019, there would also be an increase in learning how to deal with biased data. It can be argued that increased government regulations can address this issue but these regulations are often not up to speed with the technology advancements. The onus lies with the businesses to take proactive steps to adopt principles of non-discriminatory data. The World Economic Forum has released a report to prevent discriminatory outcomes in machine learning. Some of the ways this issue can be prevented is by ensuring 1.active inclusion from diversity of inputs 2.review of potential risks and 3.balancing transparency in lieu of speed or performance. 7. Neural networks: To put it briefly, neural networks or artificial neural networks emulate human brain. They store all data in a digital format – sensory, text or time and use it to classify and group the information. For example, reading someone’s handwriting comes easily and unconsciously to us whereas in order to teach an algorithm we feed it with vast amounts of handwritten data to recognize patterns in it. There is a huge demand of neural networks in robotics, to improve order fulfillment, prediction of the stock market, and diagnosis of medical problems or even to compose music! The current neural network technologies will be improved upon in 2019. This would enable this type of AI to become more sophisticated as better training methods and network architectures are developed. Neural networks are also fundamental to “deep learning”, powerful set of algorithms that can be used for image processing, speech recognition or to process natural languages. This is helpful in helping autonomous vehicles or fraud detection or even to look for signs of cancer. 2019 would continue to increase the research that goes into it so that we can take advantage of the power of these applications for life-altering situations. 8. Socio-economic models: At every AI event I organize or attend, a common question on everyone’s lips is “would AI take away our jobs?” The answer in a nutshell is “it depends”. While AI would take away mundane jobs where there is a scarcity of resources (e.g., agriculture, manufacturing or warehouse assistants), it would also enable newer jobs with different skillsets. Whatever is the answer, this topic is very popular and is under discussions by many governments, the United Nations and the World Economic Forum. This is because the onset of AI applications has the risk of widening skills gap and has potential to create polarized societies. While AI applications drive new skills and new jobs, it is important to complement it with value-creation. For example, although automation might remove the need for certain jobs, there would also be a demand for high-touch jobs such as customer service representatives, teachers, caregivers etc. Some countries like Finland are even experimenting with the concept of the Universal Basic Income. Redistributive programs would continue to be the focus of attention for lawmakers and subject to debate and discussions in 2019. 9. Deep learning: Machine learning, the most popular form of AI algorithms, becomes challenging when the number of dimensions of data increases. For example, calculating the price of a home given existing home prices in a location has only two dimensions of data. Imagine trying to transcribe your voice into text. The problem is now exacerbated a hundred times. Deep learning is also the technology behind self-driving cars, voice control as well as image recognition. With the advent of both Amazon’s Alexa and Google home, there is a wide range of voice-enabled applications that use natural language processing (NLP) algorithms, an example of deep learning. This has increased the interest in the next generation of deep learning algorithms that can solve even tougher problems such as interpreting technology infrastructure issues. 10. Privacy and Policy: The introduction of GDPR was a much-talked topic in 2018. 2019 would see further privacy and policy conversations. This is important in order to protect our privacy and ensure organizations approach data privacy earnestly. Most of us are unaware of how our digital information is being used. It is sometimes lost in the fineprint and other times not even informed about it’s usage. Facebook’s recent crisis over privacy policies is the tip of the iceberg. Countries and legislators view of privacy and policy around it would continue to be of importance in 2019. The issues of consent of usage of a system, especially around AI applications would be huge given that the laws surrounding AI is still new and needs further understanding. Countries around the world would continue to work on strategies and initiatives to guide the development of artificial intelligence (AI) Regulations. Standards that are critical to ensure that safety, transparency and awareness of the complex AI technologies would also be developed. Conclusion: In conclusion, Artificial Intelligence is not going to see a decline any time soon. The growth of it continues as we enter 2019 and the focus would not only be on new technologies and applications in the industry but also how it intersects with the society and heralds in technology for the better.