Review of the A.I. Bootcamp
Speaking and hearing, understanding and acting
Artificial intelligence (AI) is a major customer service topic for Deutsche Telekom customer service; however, we are also interested how it can be applied to other areas – above all the exchange of ideas with other AI industry innovators. This is why Deutsche Telekom invited 12 start-up companies from Israel, Canada, Poland, Switzerland, and the USA to a boot camp at the Berlin offices of its hub:raum subsidiary. There they could present their ideas and services in the fields of cognitive science, emotional intelligence, image & speech recognition, and deep learning.
Dr. Tina Klüwer opened the boot camp with a keynote speech. She is a computer linguist with years of experience in international research projects, and expertise in dialogue systems and language-enabled agents. She spoke about artificial intelligence and the difficulties involved in teaching an artificial intelligence to distinguish between similar things. To illustrate, she used several images to demonstrate how difficult it can be to differentiate between a Chihuahua and a raisin or blueberry muffin: her point being that if this can sometimes difficult for even humans, how should a machine overcome the problem? The same can apply to customer billing enquiries, with can differ greatly even when they are spoken in a similar way, “Where is my bill?” as opposed to, “When will I receive my bill?” Some languages – such as English – also have synonyms for the same word: “invoice” for “bill”, for example. Everything that humans often do intuitively, an AI has to compute.
Comprising Dr Tina Klüwer; Paul Thiekötter, a partner with FlyVentures; Fabian Westerheide, founder of Asgard Capital; and Jan Hofmann, VP eCompany Products for Deutche Telekom, the panel of experts discussed current trends and the relevance of AI. They were unanimous in their assertion that the next significant step in intelligent machine development would be a move away from manually written programs with multiple of predefined rules, towards programs that learn independently using data and interactions with humans.
However, they noted that one major problem along this path would be that most data is not available in a machine-readable format. “Pimping up” existing databases, or even digitising data, would therefore be an essential prerequisite. Many employees will be required over the coming years to carry out this task. It served as an excellent example of how digitalisation can create jobs – not just destroy them, as people often fear.
Also in the future, machines will not replace people. Jan Hofmann is convinced it will be much more the case that AI will take over standard tasks for humans, thereby allowing more time for call centre agents, for example, to concentrate on advisory duties. This is why he also believes that within 10 years at the latest, dealing with AI and the interaction between digital/automated and analogue/human intelligence will have become totally natural.
The start-ups that presented themselves at the boot camp harness the potential of their AI systems in very different ways: for example, ELLEN, the AI from the Polish Quantumlab, analyses reactions to advertising and marketing campaigns. These reactions are then processed and made available as objective data. ELLEN is already in use with various market research institutes and academic researchers. It has also been tried and tested in practical use: in restaurants, it recognised the feelings of both customers and employees; thereby making the sales process easier.
Fluent.ai Inc. is more involved with intentions than reactions. This voice-based application from the Canadian start-up recognises intentions, personalises them, and through this it learns from context, behaviour, and language – in all languages and with all accents. The interaction interface that makes this possible also be used off-line or in very noisy environments.
The founder of Fred knows IT presented a trouble-shooting service that helps customers to solve their IT problems, for example if they are connecting up a router. Simple dialogues enable users to hone in on their problem, until the AI ultimately provides the right solution. The Gigaaa chatbot is still under development, and should soon become available in the Apple App Store and Google Play Store. The developers behind Dragan Alexander Stevanovic and Thomas Geyer are promising a “Next level chatbot”.
The Swiss company Spitch is harnessing automatic speech recognition (ASR) to approach biometric voice recognition, voice user interfaces (VUI), and natural language data analysis. The company merges mathematics with customised technology adaptations.
The final start-up companies that presented themselves were already in the later stages of financing. Three of them were from the USA; two from Israel – which is home to a lively start-up scene in Tel Aviv; there was also a company from Germany, and a dual-national start-up with founders from Spain and Germany.
The first of the US start-ups to present itself was called Cogito. It aims to improve interactions, for example between call centre agents and customers. To achieve this, the AI analyses conversations and provides customer service employees with real-time support, allowing them to deal better with the caller’s needs.
Gridspace is also involved in communications between customers and companies. This start-up has branches in Los Angeles and San Francisco, and operates the leading conversational intelligence platform. Companies can use this to analyse written and spoken customer dialogues. Here too, the aim is to optimise communication and contact with the customer.
Kitt.ai proves how simple it can be to build chatbots. Its product known as ChatFlow enables dialogues, using information from a variety of sources.
Likewise, Onvego also wants to simplify working with AI. This Israeli start-up provides an end-to-end toolbox for companies and developers, allowing them to build speech-based conversation interfaces. The platform includes integrated solutions for natural language processing (NLP) and artificial intelligence (AI) technologies, so that developers no longer require any specific expertise to work in this area. They can seamlessly integrate the corresponding interfaces into their products, thereby creating user-friendly speech-controlled Apps or IoT applications.
Also from Israel, the start-up Lexifone moves beyond pure “language to text” services within speech recognition. Its real-time translations enable speakers of multiple different languages to participate in telephone conferences – much as if they had the Babel fish from “The Hitchhiker’s Guide to the Galaxy” speaking in their ear.
TwentyBN is based in Berlin-Kreuzberg. Its name stands for 20 billion – the number of nerve cells that work in our cerebral cortex and make the human brain so successful. The founders have imitated this principle, and have created an artificial brain, known as the cortex. This neural network is a structure of artificial neurons that learns things in much the same way as humans. It is being trained to recognise images, translate text, answer questions – and especially to recognise patterns.
Last but not least, the German/Spanish start-up Narrativa deals with the automatic evaluation of large quantities of data. Its bot, known as GabrieleAI, learns the context and tonality of data using Natural Language Generation. This results in texts that can, in some cases, hold their own against those generated by journalists. It is already being used for stock-market reports and football articles, with the potential for it to generate texts for every game within every league – good news for fans.
The AI boot camp made it possible for like-minded people to enjoy an enriching exchange. One of the founders noted, “It’s not just about exchanging knowledge, it’s also about networking.” Another said, “I love the informal atmosphere here at hub:raum. This is one of the most important incubators, and for me personally it is undoubtedly the most important.”