Asheem Chandna | Where Does AI Go Next?

Hi everyone. Welcome to Grey Matter, the podcast from Greylock, where we share stories from company builders and business leaders. I'm Heather Mack, head of editorial at Greylock. Today's episode is the audio version of Greylock general partner Ashim Chandna's essay entitled Where Does AI Go Next? The essay originally appeared in Forbes. You can read it there or on the greylock website@greylock.com greymatter. You can subscribe to the Gray Matter podcast on Spotify, SoundCloud, YouTube, or wherever you get your podcasts. Where does AI go next? By Ashim Chandna over the last decade, artificial intelligence has been a relentless source of business innovation. Now its influence is about to expand dramatically. Once a staple of science fiction, AI has quietly forged a critical role in some of the most ordinary yet essential business tasks. Automation of business processes, data analysis, defect detection and manufacturing, and basic interactions with customers all are an embedded part of modern business and are increasingly enabled by AI. Where, then, does AI go from here? Even as the pandemic helped accelerate the shift to cloud computing and remote work, there's also been a burst of innovation in readying AI for its next stage. Many computer scientists, economists and investors believe that we're at the threshold of enormous leaps forward in artificial intelligence and machine learning, with implications for both entrepreneurs and businesses. GPT3 as a paradigm Shift Most applications of AI today tend to involve training computers to pattern match images and data so that they can recognize new examples of them in different settings. Showing a computer hundreds of images of, say, red traffic lights or potatoes allows computer scientists to create applications that recognize these things and, where necessary, act on them. This kind of artificial intelligence is now widely deployed, and we now take it for granted. The next significant step for AI emerged during the worst months of COVID 19 with the launch of OpenAI's GPT3, which uses Transformer AI models to enable a computer to not only recognize images and patterns, but actually generate language, text and images on its own. This is a paradigm shift, says Mustafi Suleiman, CEO at Inflection AI and one of my partners at Greylock. Mustafa previously co founded DeepMind, which was acquired by Google. To date, deep learning has mostly been used for classification tasks, he says. These new AI models are capable of generating entirely new, high quality content. Suleyman explains that these transformer models help computers interact with humans, absorb their language, reference conversation, and then generate entirely new dialogue and language. These machines will help sort through, summarize and prioritize the huge quantities of information we interact with every day, he says. Yet it is a mistake, he cautions, to think about this next stage of AI as just advancing a single service, for example generating responses to questions. He believes that one of the significant achievements of emerging AI technology is that much of it will grow from low code environments. As a result, language generating AI will proliferate widely. We may soon see a world where every brand builds its own AI system to directly interact with customers, he says. AI and Business Prediction AJ Agrawal is a professor at the University of Toronto and co author of the forthcoming book Power and the Disruptive Economics of Artificial Intelligence. He believes that despite much investment in AI technology, many CEOs remain skeptical about the value. He cites a recent BCG MIT study that found that while many businesses have made investments in AI, most companies still have a long way to go to get to generate substantial financial benefits. Yet Agrawal is optimistic about AI's future impact. While the first wave of AI implementations were point solutions that deliver value by lowering the cost of existing predictions like fraud detection in banking and demand forecasting in retail, he argues that the second wave will be predicated on system solutions that, quote, transform the production process and possibly the value proposition. He draws an analogy to the early days of electricity. Years after electricity had been introduced into Cities, only about 3% of American factories were using it. When new factories realized that distributed electricity not only offered a slight cost advantage relative to fuel or steam, but more importantly allowed them to decouple the power source from the machines, they began redesigning factories and repositioning equipment in more efficient ways. Electricity usage soared from 3% to 50% in two decades. Like electricity, AI will change how people perceive what is possible with the technology. The big shift in AI, he believes, will occur when business leaders recognize opportunity not as a point solution, but as a pathway to system level innovation. When banks first started to embrace AI for fraud detection, he explains, they merely replaced one set of fraud detection statistical tools with a better set of statistical tools. Yet the work remained essentially the same. This was easy to implement. The benefits of AI were immediate and measurable. In the years ahead, he sees the most ambitious leaders rethinking fundamental questions about the risks they can eliminate. Consider indoor agriculture. Until now, commercial greenhouses were largely constrained by risks related to pests. With the increasingly reliable and affordable AI driven systems that deliver early pest detection, a company might decide to triple the size of their greenhouse or invest in a larger variety of crops. The catalyst will be the low cost of predicting pest risk. Agrawal believes a second phase of AI will see many examples where businesses from insurance to manufacturing will discover disruptive value propositions based on the fact that AI enables new ways to manage risk by employing high fidelity AI predictions. For example, he says, with a sensor that can successfully detect and predict leaky pipes, an insurance company might spend more effort helping their customers reduce the risk of water damage for homes they predict are at high risk, even if that means selling policies with lower premiums due to the reduced likelihood of a claim because they can increase the volume of policies. In other words, the next phase of AI is not merely an advancement in computing capability, but something bigger that has the potential to impact strategy, economics and business boundaries for all companies. Business strategy will increasingly be shaped by AI AI as a Business Tool Sam Moto Midi, my Greylac partner who is a domain expert around AI, sees both the power of prediction and customer interaction coming together in new uses for AI. Like Mustafa Suleiman, he believes that the ability of computers to be pre trained on immense data sets will allow the computers of the future to solve general problems and be fine tuned to take an active role in specific real time business situations. As he says, the new AI models are excellent at generating information and conversation flow on the fly and and rapidly improve from human feedback. Motimedi sees immediate implications for sales, customer service centers and any interaction between an organization and human beings. He points to Cresta, a fast growing company whose AI engines listen in real time to customers and instantly develop insights and suggest solutions or next steps. In addition to Cresta, the market has already seen other early efforts to use AI language generation to solve business problems. Jasper, for example, generates marketing copy based on limited input about a product. TextGeo does the same for creating recruiting material, developing culture change or driving digital transformation. Under the banner of augmented writing, Motamidi argues that the combination of language and prediction will make AI relevant to all business workflows. He cites cybersecurity, front office applications, IT and service management as just some of the areas where the underlying technology will soon be reshaped. With AI at the core, among the practical innovations will be the ability to read customers through video, act on IT requests or make predictions. In managing loan applications or creditworthiness. In healthcare, which has long been a technology laggard, there is an immense opportunity for AI to sort through all types of billing and service coding, an area where startup community notable health has made a mark with intelligent automation. The Enduring Challenges despite all the advances we may see with AI, the there remain some stubborn and enduring problems that will always be a part of the technology. In a conversation last fall with Greylock's Reid Hoffman. Fei Fei Li, the co director and co founder of Stanford's center for Human Centered AI, argues that deploying sensors in all aspects of human life raises new questions. As she says, we are excited as technologists to think about how computer vision and smart sensors and edge computing can help. But we are also confronted with the question of privacy, which with the question of legal ramifications that we never thought of. What if the censor picked up Caribbean cases? Can they serve as legal witnesses or some other adversarial events? Steve Johnson, in a recent essay in the New York Times Magazine, explored more of the controversies about computers that are capable of generating full, lucid and original paragraphs on any subjects as if they emerged from the mind of a literate human being. He raises the potential for bots to generate disinformation that appears to be authoritative. Mustafa Suleiman recognizes the same dilemma. While AI bots can be good at detecting misinformation, he says, they may also be better at disseminating it at a large scale. Sam Modamidi sees these issues as part of the inevitable regulatory oversight that AI will have to accept. All of us will have to confront questions of bias and fairness when we deploy AI to make decisions, he says. For example, what if someone is denied a loan as a result of a machine learning algorithm? He points to Trueera as a company that is helping to address these questions. In this context, AJ Agrawal offers a useful reminder despite advances in computing power, AI remains a tool about prediction, not judgment. Judgment is what humans must still do with the predictions that computing serves up. That remains a good guide as we prepare for more AI in our future. That future is coming quickly. The next era of enterprise AI will offer much faster time to value. No one doubts that we have seen only the tip of the AI iceberg, operating at much higher efficacy and automatically generating original content, not just recognizing patterns. Democratizing prediction with the potential to disrupt and enable new business models and an explosion of new AI enabled applications will be the hallmarks of the next wave of AI. That concludes this episode of Gray Matter. If you're not already a subscriber, you can sign up on Spotify, SoundCloud, YouTube, or wherever you get your podcasts. I'm Heather Mack. Thanks for listening.