Matt Lythe, Managing Director
Over 500 industry leaders, academics, users, government officials and other interested parties gathered at Auckland’s fantastic ASB Bank Waterfront Theatre last week to discuss the advancements in AI and how they are shaping our future.
The meeting included a wide range of speakers covering important research areas such as; language processing and translation, facial recognition, digital assistants, conversational AI, computer vision and robotics. Sprinkled throughout the technology and applied AI presentations were discussions on data, privacy and ethics and, following the terrible events of March 15th, a panel discussion on whether or not AI can help create a safer society.
For Lynker Analytics Principal of Data Science, David Knox and I - it felt like we were still at or near the start of an amazing revolution in this technology and that we are only really starting to unravel what it can deliver right across the economy.
This is not an exhaustive review of the conference but my reflections on the key learnings and state of play at this point in the evolution of this very impactful technology and what it means for the engineering services, environmental sciences and geospatial community. My summary falls into four key themes which are very topical right now in AI.
1. Machine translation
In a remarkable presentation one of the largest AI companies in China, IFlytek presented to us for over 20 minutes in Chinese. This presentation was processed, converted into text and then translated in real time before being presented back as text to the audience. There was a very short lag of perhaps 10 seconds during which the AI improved and corrected grammar and sentence structure.
This and several other presentations showed that the processing of spoken language and translation from one language to another is essentially a solved problem. The focus now is on natural language understanding and providing context to machines to help them process and translate between languages.
2. Computer Vision
Computer vision is the field that deals with how machines can gain high level understanding from images or videos. Until recently it was viewed as a task that separated humans from machines. Several speakers from NZ and abroad showed that computers in fact probably now exceed a human’s ability to achieve visual understanding. Tom Lintern from Homes showed how they are classifying and tagging property photos using AI to improve the process of searching for property. In a fascinating presentation Professor Jingyi Yu from ShanghaiTech University demonstrated their intelligent vision research including technique to infer faces and body shapes in occluded images.
Finally, Sean Gourley from Primer.AI, a Kiwi now based in San Francisco gave an illustration of how far Generative Adversarial Networks (GANs) have come in only 5 years. A GAN is a class of unsupervised machine learning where two or more neural networks compete in the generation of content and the detection of generated versus “real” content; in this case realistic images of human faces. The result is generated images that look authentic to human observers. Take a look at the outputs in this image and you be the judge!
Seeing this progression in the technology reaffirmed our focus at Lynker Analytics on this sub-field which we believe can deliver new and significant gains to the engineering services and environmental sciences domain.
3. Conversational intelligence
This topic received a lot of attention reflecting the increasingly wide adoption of things like Siri, Cortana, Alexa and others. Two presentations stood out. Firstly, Ed Liebenberger from Jade introduced us to Ace – a digital employee who after asking the audience 5 or 6 questions like age, occupation etc generated a life insurance quote. These conversational agents can quickly reduce time to value for organisations looking to improve customer interfaces.
Then the founder and CEO of FaceMe, Danny Tomsett, chatted to the chatbot now in new BMW 3 series vehicles - describing how the technology is moving from tech as a tool to tech as an actor with increasingly humour, empathy, emotion and companionship.
4. Ethics
This was the topic that received the most attention with several speakers as well as two panel discussions traversing the ethics of AI. Because we are the first generation to bring AI to society, we have important obligations to get it right. Adam Cutler from IBM made the point that we are responsible for teaching machines what we value while Dave Heiner from Microsoft pointed out that values in the east and west differ, so an AI needs to reflect its audience of users.
There was some very positive discussion about how AI is liberating for people with disability or creating better outcomes for people understanding health or medical advice through conversation. A key point was that diversity in training data is critical as AI becomes more widely used. The example of propaganda bots used by both sides in the 2016 US presidential election was given as a warning of how AI can negatively influence. It was interesting and positive to see that both these two big tech companies have ethics panels and codes of conduct in advanced stages.
Summary
Overall AI-Day19 – the second such event – was highly valuable. The industry is moving rapidly into the cognitive era – one where AI can make judgments and form hypotheses based on incoming data streams while also learn and adapt decisions and line of thinking. We saw several examples of this already working. The computer vision and natural language sub-fields have received significant research funding and are in an increasingly mature state enabling a broad range of industrial applications. This is relevant to the engineering services and environmental sciences sector where geospatial data, time series information, satellite imagery, drone, video and IOT sensors are prevalent.
Another key learning that is important to us at Lynker Analytics also is the need to explain the algorithmic decision making used by an AI. It’s not good enough to just say ‘magic happens’ – even if deep learning is difficult to describe, an outline of the decisioning process is important to maintain transparency in AI.
Finally, Sean Gourley from Primer.AI made a salient point when analysing the use of AI in medical diagnosis. He said the best result was not human or AI but was achieved through the combined interpretation of the data by the trained physician and the AI. He went on to say:
“We need to combine the best of human intelligence with the best of machine intelligence to create the best intelligence we can achieve”.
We are pretty excited by the potential of this burgeoning technology and look forward to showing the industry how geospatial data when combined with AI can bring new intelligence to your organisation. Roll on 2020.