Cengine blends human and artificial intelligence together to create CensTM -

EFTA00593086 Dataset 9 14 pages Download original PDF Download as text
Cengine Cengine blends human and artificial intelligence together to create CensTM - (Intelligent Avatars) and uses real time communication to make the worlds first Al marketplace. Confidential © Cengine 2016 1 EFTA00593086 Founding Team James Tagg James Tagg is a serial entrepreneur specializing in man/machine interfaces and communications. He was instrumental in developing the first touchscreens and built the first mobile VoIP and Messaging products for smartphones. James Tagg made the first ever IP mobile phone call on an iPhone at Demo in 2008. He holds over 200 patents and has written extensively about the relation between human intelligence and computer intelligence. Erik Viirre M.D. Ph.D. Erik Viirre M.D. Ph.D. is a Professor at the University of California, San Diego (UCSD) in the Departments of Neurosciences, Surgery and Cognitive Science. His scientific interests include vision, hearing and the vestibular system and higher cognitive function. He has published a broad range of research and is currently researching AI medical diagnosis. Dr Viirre led the medical team that enabled Stephen Hawking to experience weightlessness. Confidential © Cengine 2016 2 EFTA00593087 Problem • Artificial Intelligence (AI) often struggles with problems that Human Intelligence (HI) finds relatively easy. • AI fails catastrophically when it exceeds one of its limit, yet it is often not aware it is struggling! • AI failure can be frustrating - customer support systems, or dangerous - medical diagnosis. Confidential © Cengine 2016 EFTA00593088 Solution • When an AI nears its cognitive limit - the boundary of its known domain - we need a way to tell if it is about to fail and allow humans to seamlessly intervene. • Our technique detects when we need to move from an application of AI to the use of HI in real time and preemptively. Confidential © Cengine 2016 4 EFTA00593089 Technology • We are developing a mathematical tool to detect when an AI is about to fail, but training such a system is difficult. • We have realized a practical shortcut exists using communication tools to bring humans into the mix based on the measure of empathy and engagement. • Using multi-mode interaction we can train AI systems using humans, gradually reducing the need for human participation. • For systems that will always need a human for safety and ethical concerns, this builds a Human Artificial Intelligence hybrid, a CenTM. Confidential © Cengine 2016 EFTA00593090 UA LE Interface Moderating Human allF" ti metft, Predicted probability of Intervention Confidential © Cengine 2016 EFTA00593091 Interface Activity Human.Jane Avatar.Fred AA • A human monitors the interaction between a human user and an avatar. They can see video, hear audio and see the natural language parser text generated by the AI in original language or translation. • The moderating human has information on the logical complexity of the transaction and the emotional empathy in the transaction. The emotional empathy includes tone of voice as well as face and body language cues. • A single human can monitor many transactions. When a transaction looks like it is failing they can step in. • This creates a training set. Confidential © Cengine 2016 7 EFTA00593092 Human Intervention Types • Humans can override at a number of points • Before the response, using the mathematical predictor • After the response, using the emotional response of the user (not so good) • Human can override • In Background • Answer as the avatar • Improve the model or data • Step into the foreground • Answer as Manager 'Let me get my manager' • Consultant (in medical context) UA Confidential © Cengine 2016 EFTA00593093 Example: Let me get my Manager AA • If the task gets too complex and the rapport has broken down between Avatar and Human, the AI can bring their 'manager' in. • One of the moderating humans that has been monitoring a set of conversations comes into the chat and replaces the AI for a while. • The User interface keeps track and redistributes the work amongst other human assistants. Confidential © Cengine 2016 9 EFTA00593094 Human Skill Marketplace • Moderators join the human skills unit to form a community of suppliers rather like Airbnb, although the skills are far more specific. • These moderators train the AI and catch errors. • The more they train, the more 'blocks' they earn, giving them a share in future revenue generated by the AI version of themselves. • They are also paid for error correction work and providing oversight. • For many tasks human input is always needed for reasons of ethics, safety, oversight and the assumption that AIs will always struggle with new scenarios. • HI is different to AI and our aim is to make them work well together. Confidential © Cengine 2016 to EFTA00593095 Applications First Products • Personal Friend (Chat bot) • Very first product is just a friend (character to help lonely people. In app purchases may provide some monetization.) • Medical counselling chat bot is the first paid product. • Performs the job of a clinician force multiplier, allowing one doctor to see multiple patients at the same time. Further Applications • The technology can be applied to low risk applications first • Personal helper • Customer service • Low intensity medical diagnosis • The training should generalize to higher risk applications • Driverless cars • Higher intensity medical diagnosis Confidential © Cengine 2016 11 EFTA00593096 Differentiation & Value creation • We are building an AI / HI hybrid able to interact with problems in real time. • We will build a large data set of situations where an HI/AI decision needs to be made, as well as the algorithms to select appropriate reactions. • Participation in this HI / AI hybrid is on an open basis rather like Airbnb. Moderators log in to add their HI to an AI task. We build community and user /supplier base. • We are also targeting more specialized problems such as medical diagnosis and counselling which involve high $ per hour values ($1000-$1500). Rather than more general problems. Confidential © Cengine 2016 12 EFTA00593097 Finances • We are seeking an initial investment of £5m c. $6.25m. • The funds will be used to build and launch the communications AI HI market place, and ready the product for trials in medical counselling / diagnosis assistance and training. • One trained doctor could simultaneously handle up to 4 patients using our system. Given costs run between $1000 and $1500 per hour for this sort of professional there is a large opportunity. There is also huge shortages of skills in these areas and growing demand. Confidential © Cengine 2016 13 EFTA00593098 Contact James Tagg CEO Confidential © Cengine 2016 14 EFTA00593099

📷 Images in this document (14 detected; 6 largest described)

AI-generated factual descriptions of embedded images (llava:13b). These are searchable across the corpus.

[Image 1] The image appears to be a screenshot of a computer interface, possibly related to a speech recognition or voice analysis system. There are several photographs of individuals displayed, each with a name and a timestamp. The names are "Cen Fred," "Human Jane," and "Human Cen." The photographs show different individuals, and there are audio waveforms and possibly transcripts or scores next to each ph [Image 2] The image appears to be a slide from a presentation, focusing on the topic of "Interface Activity." The slide contains text and images. On the left side, there is a photograph of two individuals, a man and a woman, who seem to be engaged in a conversation or presentation. The man is wearing a suit and tie, while the woman is dressed in a more casual manner. On the right side, there is a diagram il [Image 3] The image appears to be a slide from a presentation, with a title that reads "Example: Let me get my manager." The slide contains bullet points with text that seems to be discussing the challenges of managing human-AI interactions. There are two photographs of individuals on the slide, one on the left and one on the right. The left photograph shows a woman with dark hair smiling at the camera, and [Image 4] The image appears to be a screenshot of a presentation slide or a document with text and images. The text is too small to read clearly, but it seems to be related to a founding team or a group of individuals involved in a project or organization. There are two images of men, presumably the founding team members, with their names listed below the images. The names are James Tagg and Erik Vieze. The [Image 5] The image appears to be a slide from a presentation, specifically focusing on "Human Interaction Types." The slide contains text and a graphical representation. The text describes different types of human interaction, such as "Humans can overrule a number of points before the response using the emotional response of the user not so good." It also mentions "Backgroun" and "Answer as the avatar." Th [Image 6] The image appears to be a screenshot of a presentation slide with a title that reads "Problem." Below the title, there are three bullet points with text that discusses the limitations of artificial intelligence (AI) and human intelligence. The slide includes a photograph of a man and a woman, who seem to be in a professional or corporate setting, possibly attending a meeting or conference. The man