The state of Artificial Intelligence, a summary of StartSummit
There was a great number of Artificial Intelligence related talks and discussions on the StartSummit event. Even if we don’t realize, Artifical Intelligence (AI) is part of our everyday life today. It will affect many industries and areas in the upcoming months and years. The biggest open topics are related to regulations and ethics. How can we as a society make sure that any technology which is developed and used will be used in an ethical way? How could engineers and regulators make sure that any institutions or corporations who are developing AI technologies follow clear guidelines on how to avoid programming bias into the algorithms.
How could we make sure that if we reach the point in time when a superintelligence is created it will not decide to kill humans as it could identify us as a cancerous species of the planet Earth? We need to make sure that we are prepared for the changes and disruptions by the technology. Regulators need to speed up to cope with new changes every day. More and more discussions are needed and education, information is crucial. Currently, I’m reading the book Superintelligence: Paths, Dangers, Strategies written by Nick Bostrom which is considered a basic read related to the dangers of AI. Use cases for AI discussed: autonomous cars, arts, and diagnostics.
The State of the Art – AI and what is yet to come – Keynote
Thursday 21.03.2019 11:30 – 12:00
Chief Scientist SalesforceFounder MetamindRichard Socher studied deep learning, natural language processing, and computer vision and got his Ph.D. degree. He is the Chief Scientist at Salesforce, founder of
What is natural language processing?
Natural language is the way humans communicate: speech, texts, signs, etc. Language is hard to learn, it takes years for a child to learn it. The beauty of the language is that it can be highly ambiguous. Our languages contain many nuances, changes in tonality could change the meaning completely. It is constantly evolving, different groups use a different set of words and pronunciation might depend on location. It is wonderful that most of the time we can clearly understand each other, although still there are many misunderstandings amongst humans. Defining formal rules to break down and clarify the meaning of each sentence is a daunting task. Natural language processing on one end could mean a simple calculation of word frequency to categorize writing style, on the other end it could mean the understanding the text (or speech) to be able to answer a question. Computational Linguistics and Statistical Natural Language Processing are both modern studies of Linguistics.
Mr. Socher mentioned the fourth industrial revolution, the age of intelligence, AI. He thinks that the basics of programming languages should be known by all of us. Natural language processing is used for example in Gmail, where buttons are shown to send quick answers. Amazon is using AI for recommendations and computer vision is used quite often in various social media platforms. The challenge is to understand large volumes of unstructured data i.e. vision, language, voice. Salesforce is now developing and researching one state-of-the-art artificial intelligence to understand the previous factors. One example for analysis: automate the counting of a company’s logos on the photos of an event. Another use case for AI is to estimate the damage of a car with an application published by an insurance company.
Natural Language Processing covers different areas at Salesforce. Understanding intent, sentiment / classify data and multi-task NLP which uses one engine to cover understanding intent and classify the data to be able to answer the questions. Salesforce is using automated speech recognition and they developed an autonomous voice-powered customer service. One use case for this technology is the automation of an assistant by simply asking the computer assistant by voice to organize a new meeting and handle follow-ups. As a real-world scenario, Salesforce provides an automated sales engine which can automatically engage in two-way conversations, handle e-mail and SMS in multiple languages. Many companies already use automated systems if you call their support. I expect this technology to be extended and used by many more companies in the following few years. Imagine a world with zero waiting time on support lines, as machines could take an unlimited number of calls. If the systems which are being built will be efficient then most problems could be solved in a very efficient way. I’m a bit skeptical as profit-oriented companies might not put a lot of effort on developing their support and quite often if you face a problem which doesn’t affect many users your problem might be ignored or neglected.
AI use case in Medicine is a highly promising area. Counting white/red blood cells in blood used to take 3-4 days costing 400 USD. With AI it could take 60 seconds and cost less than 40 cents. In radiology, we’ve already seen great results as AI is capable of outperforming the performance of doctors. AI might be used to automate answering of questions of patients: check symptoms, assess risks, analyze patient data, discuss nutrition, etc.
AI in agriculture can already water plants in an automated way and even more importantly: it can recognize which plants need fertilizer and only spray fertilizer or pesticide on those. This is clearly a win-win situation. Less cost for the farmers, and a healthier environment, as they don’t use so much fertilizer and pesticides which might end up in the water system. The challenges here might be how to support small farmers with the same technology.
AI use case: protect beaches from sharks by flying drones over the area and using image recognition.
In my opinion one of the most important aspects of AI is bias. The algorithm is as good as the input data. If the data is biased, in a credit application the credit classifier algorithm could become sexist or it could prefer certain races or characteristics. The solution is to make sure the data is representative.
The tips from Mr. Socher to make the world a better place: develop your skills, write down and revisit your values and switch between focused work and high-level thinking and questioning.
I’ve asked Mr. Socher on how we could enforce unbiased data. It is complicated, however, needs careful organization and handling of the data. Unrelevant data could be left out from the input and thus could lower the bias.
AI – Hope is not a Strategy – Keynote
Thursday 21.03.2019 13:45 – 14:15
AI is a general-purpose technology that will bring about disruptive changes in the economy, society, and politics. How can we strengthen the Swiss start-up scene? How can we manage negative externalities through intelligent regulation? How can we create common values and ethical foundations that enable us to develop and apply AI for the benefit of all?
National Technology Officer Microsoft Switzerland
You have to be an acting part of this AI revolution. That’s the condition of having a say in designing and defining the rules of AI.Emmanuel Macron
Microsoft is emphasizing the importance of discussion of AI between shareholders, the private second and academia. Microsoft is pushing for a governed AI to achieve collective, conscious decisions
Switzerland is well positioned to actively shape the AI development and there is a growing number of AI use cases implemented. Trust needs to be achieved which means meeting the needs in terms of privacy, security, and safety.
Mr. Holtischer mentions a collaborative project between Microsoft and Bühler group. They have developed hardware called LumoVision which provides a solution to remove grains which contain cancer-causing toxins. Special cameras and sensors enable LumoVision sorters to reduce aflatoxin contamination in maize by up to 90% with a yield loss of less than 5%.
AI is helping automated predictions based on data. The prediction is not the decision making itself, it is an input for the decision making. Decision making consists of judgment and action. Human judgment can’t be replaced by a machine. He mentions empathy, emotional intelligence, creativity. Human intelligence should be augmented rather than replaced. Human judgment should be built into AI systems.
Microsoft is mostly skeptical about creating closed-loop systems when AI systems would decide and act without human intervention (automated decision making). A critical use case might be social welfare decisions or deciding who gets medical treatment. At the minimum, there should be a way to understand how a decision was made by the AI. What influenced the decision and what could be changed to modify the result. This is called explainability of AI. For example, there would be a need for understanding if an AI would define someone as critically sick. The patient and relatives would need to understand the reasons behind this diagnosis and proof is required.
Computers are neither ethical or non-ethical. It is the responsibility of the companies and startups to design and deploy the solutions in a way that they are following certain principles. Microsoft defined six principles for Ethical AI.
Accountability, Transparency as basis and Fairness, Reliability & Safety, Privacy & Security, Inclusion.
Microsoft created the Aether Committee, where “Aether” stands for AI and Ethics in Engineering and Research. This committee might turn down certain sales, if they think the AI solution might have a negative impact on human rights or if ethics are in question. Indeed the committee already turned down some sales, although Microsoft doesn’t release details. Some critical areas are military applications and applications in which AI would be for face recognition or predictions related to this. Both areas are not supported by Microsoft. In an interview with Tobias Steger mentioned that they are working with the US Military in terms of defense applications of the AI systems. Facial recognition might be used to help families reunite in case of disasters or during wartime, it could be used to identify terrorists in a soccer game and it could be used for tracking of individuals by governments, even without approval. This shows that technology can be both used or misused. Suggestions for startup building or using AI technology Microsoft published guidelines. Diversity in the development team is required and developing a code of conduct, ethical guidelines are suggested.
Technology development doesn’t just happen – it happens because us humans make design choices.Satya Nadella
Microsoft thinks that private sector self-regulation is sensitive areas is not an option, as one company might change their practices, the others might not follow. Dynamics of competition between companies and especially between countries might result in companies implementing AI and facial recognition in a way which would not be accepted in many countries. Microsoft thinks that powerful technologies such as facial recognition should be governed and discussions should happen between private and public sectors, policymakers, society. AI must be developed in a human-centered way. Mr. Holitscher thinks that more discussions and reflections would be needed about these topics. Microsoft will lunch a Swiss-based cloud system with two data centers (Zurich, Geneva) in a couple of months. Customer data handled in these centers will not leave the country at any point. Microsoft made a workshop on 22nd of March at the StartSummit. They aim to develop a call to action on how to deal with AI in Switzerland. The results of the discussion will be presented to an expert team which prepares a report of AI which finally will be presented to the Swiss Federal Council in September of 2019.
Creative Superpowers – How AI will power the next Generation of Artists – Keynote
Friday 22.03.2019 14:45 – 15:00
The competition for the lead in AI has significantly increased and for instance, caused Chancellor Merkel to call for an increase in investments in this particular field. But where does Europe in comparison to the USA and China actually stand and how can we ensure that it can keep up? Florian Meissner will elaborate these questions by stating examples, strategies and own experiences he had with his company EyeEm.
Florian Meissner shows fake war photographs, mentions that facial recognition systems could recognize people with scarfs or hats. He shows the capabilities of AI by showing two photos with different styles merged into one, and the background or watermark of photos being removed by algorithms. He showed a series of generated portraits based on a couple of photographs. The end results are stunning, no one would be able to tell that the results are not real photographs. He also showed the fake video with Obama shown below.
This brings up the question of how to differentiate between fake and real photos and videos. The next question what Florian Meissner asks is:
Can AI create art?
He mentions the painting generated by Artificial Intelligence which was sold by Christie’s for $432,500, beating estimates of $10,000.There are interesting questions to be answered here: who is the real artist here. Is it the group Obvious (Pierre Fautrel, Hugo
AI supported medical Diagnostics – Are the Doctors ready? – Keynote
Friday 22.03.2019 15:15 – 15:30
Founder & CEO
Peter Ruff starts his presentation by mentioning the importance of personalized therapy and empowerment of patients. The areas where AI is already used: radiology, drug research, diagnostics, triage, nursing. His company focuses on avoid heart attacks as this is a cause of a major number of deaths each year. The underlying cause is stenosis which means that the coronary arteries are getting narrowed because of fat, calcium, etc. If the shell of these plaque cracks a blood clot will form, and if this clog totally blocks the artery, it leads to permanent damage to the heart muscle cells. When this happens it needs to be treated immediately. Their product Cardio Explorer is a non-invasive Artificial Intelligence-based
Artificial Intelligence and similar technologies like the blockchain will disturb all industries. The question remains open: how do we handle these disruptions? Can we face and solve the challenges of regulations, education, and ethics? I’m positive, however, not naive. This challenge is