At Rise of AI 2019 in Berlin, Wolfgang Hildesheim, Director of Watson, Data Science and Artificial Intelligence DACH at IBM, walks through client projects IBM ran in Germany, Switzerland and Austria over the previous two years, opening with Deep Blue's win over Garry Kasparov and IBM Watson's 2011 Jeopardy victory as milestones before today's systems. He shows Project Debater, an IBM system that argued both sides of a policy question against human debate champion Harish Natarajan, then organizes commercial AI into four capability groups: conversational assistants, visual recognition, document understanding, and speech-to-text. Case studies include a dental insurer's chatbot that reportedly doubled sales, Austrian startup Anyline's license-plate recognition, a multi-year in-car assistant built with Mercedes-Benz covering more than 300 topics across several model lines, a document-understanding tool that finds expiring contracts among millions of PDFs, and CIMON, a floating AI companion robot developed with the German Aerospace Center and Airbus for the International Space Station. He closes on IBM's Watson OpenScale tool for monitoring AI fairness and transparency, IBM's GDPR-compliant Frankfurt data center, and practical advice for companies starting AI projects: move fast in small steps, invest in data and people, and pick partners who protect data and intellectual property.
Key takeaways
- Hildesheim frames the current AI wave through the Gartner hype cycle, warning companies that disengage after early disappointment often lose to competitors who kept building.
- IBM's Project Debater argued both sides of a policy resolution against human debate champion Harish Natarajan, extracting pro and con arguments automatically.
- A dental insurer's chatbot reportedly doubled policy sales, and Austrian startup Anyline's visual recognition business was doubling year over year.
- The Mercedes-Benz in-car assistant, built with Daimler, covers more than 300 topics across several model lines and includes augmented reality guidance.
- CIMON, an AI companion robot built with the German Aerospace Center and Airbus, assisted astronaut Alexander Gerst on the International Space Station and can detect up to twelve emotional states in a person's voice.
Detailed notes
From Deep Blue to Watson: setting the scene
Hildesheim opens with Deep Blue's victory over Garry Kasparov as an early marker of machines outperforming humans at complex tasks, noting a chess engine once impossible now runs as a free app. He then shows footage of IBM Watson's 2011 Jeopardy win against human champions, calling it a breakthrough in natural language processing and question answering, and stresses that systems described as "intelligent" only appear that way from the outside while being narrow and "dumb" internally. He places the current wave on the Gartner hype cycle, warning that after peak hype many organizations conclude a technology is too slow or expensive, disengage, and then lose ground to competitors who kept building through the trough.
Project Debater: arguing both sides
Hildesheim shows a demo of Project Debater, an IBM system that argued for and against the resolution that governments should subsidize preschools, facing human debate champion Harish Natarajan. He presents it as evidence that AI systems can now extract arguments for and against a topic and go deeper into semantics, while cautioning that this is applied capability, not general artificial intelligence, and that companies underestimate the opportunity in adopting what already exists.
Four building blocks: read, see, know, hear
Hildesheim organizes IBM's commercial AI services into four groups: conversational assistants (Watson Assistant), visual recognition, knowledge generation and navigation (Watson Discovery), and speech-to-text. As examples, a dental insurer built a chatbot, the "empathic insurance assistant," that its board says sells more than 100% more policies than before; Austrian startup Anyline uses visual deep learning to read license plates and documents and says its business is doubling year over year; and Watson Discovery's "smart document understanding" can answer questions about specific fields inside large sets of PDF contracts, for example finding which of a telecom operator's contracts are expiring within the week among millions of documents.
The Mercedes-Benz in-car assistant
IBM's multi-year project with Daimler built an in-car virtual assistant covering the A, E, C and S class Mercedes-Benz models, letting drivers ask questions in natural language, such as what a dashboard light means or how to save fuel, across more than 300 supported topics, with augmented reality integration shown for the first time in a Daimler digital product. Hildesheim says the project competed against Google, Amazon and Microsoft, and that Daimler was running up to 20 projects on the platform at the time of the talk, describing it as only the beginning of the company's digital transformation.
CIMON: an AI companion on the International Space Station
Hildesheim presents CIMON, a floating, autonomous AI companion robot built with the German Aerospace Center and Airbus, developed by a small team in under two years and deployed to the International Space Station, where it worked with German astronaut Alexander Gerst. He describes CIMON's cameras as its eyes, its Watson Assistant backend as its brain, and multiple microphones with speech-to-text as its ears, letting it recognize a crew member's face, follow them in zero gravity, and assist with biological and chemical experiments. A newer version can detect up to twelve emotional states in a person's voice and adjust its own tone using Watson Tone Analyzer, for example offering encouragement if a crew member sounds low.
Platform, trust and success factors
Hildesheim describes the IBM Cloud platform behind these projects, including Watson Studio as the AI development environment and Watson OpenScale, a monitoring tool meant to open up the AI "black box" by tracking statistical errors and checking algorithmic fairness across groups such as men and women. He says IBM sits on a German government board developing ethical AI design criteria, feeding into a proposed European Commission label for trustworthy, transparent AI "made in Germany," and notes IBM's Frankfurt data center meets GDPR requirements that some regulated industries need. He closes with advice for companies starting AI projects: move quickly in small steps rather than over-planning, collect data, invest in people, and pick a partner who protects both data and intellectual property and offers ready-to-use software rather than requiring everything to be built from scratch.
Quotes
This is really a different type of work than when I was a PhD.
Wolfgang Hildesheim
You need to collect data, you need to really invest, hire people and pick a partner.
Wolfgang Hildesheim
Questions answered in this talk
What was IBM's Project Debater and what did it do?
According to Wolfgang Hildesheim at Rise of AI 2019, Project Debater is an IBM system that argued both for and against the resolution that governments should subsidize preschools, facing human debate champion Harish Natarajan, demonstrating an AI system's ability to extract and present arguments on a topic.
What is CIMON and what does it do on the International Space Station?
Wolfgang Hildesheim describes CIMON, at Rise of AI 2019, as a floating AI companion robot built with the German Aerospace Center and Airbus that worked with astronaut Alexander Gerst, recognizing his face, following him in zero gravity, assisting with experiments, and detecting up to twelve emotional states in a person's voice.
How did IBM build trust and transparency into its AI systems for European clients?
Wolfgang Hildesheim explains at Rise of AI 2019 that IBM's Watson OpenScale monitors deployed AI systems for statistical errors and fairness across groups, and that IBM sits on a German government board developing ethical AI design criteria feeding into a proposed European Commission trust label for AI "made in Germany."