By AI Trends Staff
IBM Research AI has released its five AI predictions for 2020, in a research blog post from Sriram Raghavan, VP of IBM Research AI. He identified three themes that will shape the advancement of AI in 2020: automation, natural language processing (NLP), and trust. More automation will help AI systems work more quickly and for data scientists, businesses, and consumers. NLP will play a key role in enabling AI systems to converse, debate, and solve problems using everyday language. “And with each of these advances, we’ll see more transparent and accountable practices emerge for managing AI data, through tools ranging from explainability to bias detection,” Raghavan stated.
Among the five IBM Research predictions:
AI will understand more, so it can do more: The more data AI systems have, the faster they will get better. But AI’s need for data can pose a problem for some businesses and organizations that have less data than others. During the coming year, more AI systems will begin to rely on “neuro-symbolic” technology that combines learning and logic. Neuro-symbolic is the ticket to breakthroughs in technologies for NLP, helping computers better understand human language and conversations by incorporating common sense reasoning and domain knowledge.
AI won’t take your job, but it will change how you work: The fear that humans will lose their jobs to machines is unjustified. Rather, AI will transform the way people work, through automation. New research from the MIT-IBM Watson AI Lab shows that AI will increasingly help us with tasks such as scheduling, but will have a less direct impact on jobs that require skills such as design expertise and industrial strategy. Employers have to start adapting job roles, while employees should focus on expanding their skills.
AI will engineer AI for trust: To trust AI, the systems have to be reliable, fair, and accountable. Developers need to ensure that the technology is secure and that conclusions or recommendations are not biased or manipulated.” During 2020, components that regulate trustworthiness will be interwoven into the fabric of the AI lifecycle to help us build, test, run, monitor, and certify AI applications for trust, not just performance,” Raghavan predicted. Researchers will explore the use of AI to govern AI and to create trust workflows across industries, especially those that are heavily-regulated.
Pytorch Founder Sees Model Compiler Advances
Soumith Chintala, director, principal engineer, and creator of PyTorch, offered some 2020 predictions in an account in VentureBeat. He expects “an explosion” in the importance and adoption of tools such as PyTorch’s JIT compiler and neural network hardware accelerators like Glow. “With PyTorch and TensorFlow, you’ve seen the frameworks sort of converge,” he stated. “The reason quantization comes up, and a bunch of other lower-level efficiencies come up, is because the next war is compilers for the frameworks — XLA, TVM, PyTorch has Glow, a lot of innovation is waiting to happen,” he said. “For the next few years, you’re going to see … how to quantize smarter, how to fuse better, how to use GPUs more efficiently, [and] how to automatically compile for new hardware.”
Thus more value will be placed in 2020 on AI model performance and not only accuracy, how output can be explained and how AI can reflect the society people want to build.
Celeste Kidd, a developmental psychologist at the UC Berkeley, says 2020 may spell the end of the “black box” references to neural networks inability to explain themselves. She predicts the end of the perception that neural networks cannot be interpreted. “The black box argument is bogus… brains are also black boxes, and we’ve made a lot of progress in understanding how brains work,” she stated.
Kidd and her team explore how babies learn, seeking insights to help neural network model training. From studying baby behavior, she sees that they understand some things and that they are not perfect learners. “Human babies are great, but they make a lot of errors,” she stated. “It’s likely there’s going to be an increased appreciation for the connection between what you currently know and what you want to understand next.”
Anima Anandkumar, machine learning research director at NVIDIA, sees more advances coming in text generation, noting that in 2019 text generation at the length of paragraphs was made possible, an advance. In August 2019, NVIDIA introduced the Megatron natural language model, with 8 billion parameters, believed to be the world’s largest Transformer-based AI model. She looks forward to seeing more industry-specific text models. “We are still not at the stage of dialogue generation that’s interactive, that can keep track and have natural conversations. So I think there will be more serious attempts made in 2020 in that direction,” she stated to VentureBeat.
She sees this next advance as a technical challenge. “The development of frameworks for control of text generation will be more challenging than, say, the development of frameworks for images that can be trained to identify people or objects,” she stated.
IT Seen Getting Better at Measuring AI’s Impact
Among trends to watch in 2020 cited in an account in The Enterprisers Project, is a prediction that IT leaders will get real about measuring AI’s impact. A new MIT AI Survey showed fewer than two of five companies reported business gains from AI in the past three years. Given the investments being made in AI, more emphasis will be put on measuring results.
Fewer than two out of five companies reported business gains from AI in the past three years, according to the MIT AI survey. That will need to change in the new year, given the significant investment organizations are continuing to make in AI capabilities. Measurements will be attempted on gains in ease of use, improved processes and customer satisfaction. “CIOs will also need to continue to put more of their budgets against understanding how AI can benefit their organizations and implement solutions that provide real ROI,” stated Jean-François Gagné, CEO and co-founder of software provider Element AI, “or risk falling behind competitors.”
Read the source posts from IBM Research AI, VentureBeat and The Enterprisers Project.