Missed our fantastic 4th Meet AI session? No problem, we’ve got you covered!
Our Meet AI series is already creating great waves! It has been featured in the Financial Times China and has attracted fascinating calibre of speakers thus far from the likes of Facebook, Microsoft, Google, Jibo, Emotech, the University of Edinburgh, Imperial College London, the University of Oxford.
Launched in collaboration with the award-winning robotics startup Emotech, Meet AI is a monthly event and AI knowledge sharing platform. Designed to celebrate London’s vibrant AI community, Meet AI welcomes AI veterans and novices alike to meet and exchange their latest insights on AI, robotics hardware engineering and product design.
Past events saw expert researchers shed light on topics such as TTS, Machine & Reinforcement Learning, and the connection between AI, Art and Animation.
For the fourth series we once again brought together London's vibrant community of AI enthusiasts, robotic developers, engineers, designers and tech lovers.
Meet AI 4 provided insight into 2 topics of particular relevance to AI entrepreneurs and researchers: 'Crowdfunding and AI' by Joel Hughes, Director (Europe & UK) at Indigogo and 'Machine Learning for Separating Style and Content' by Rory Beard Phd (University of Oxford). The Q&A that followed raised lots of interesting questions, some of which we've collected as food for thought at the end of this post.
Joel Hughes, Indiegogo: "AI and Crowdfunding make the perfect match!”
Joel Hughes, Director for UK & Europe at crowdfunding platform Indiegogo, started off the evening with an advocacy for the great symbiosis between crowdfunding and AI, quoting entrepreneur Sanjay Choudhary"Why AI and Crowdfunding make perfect partners".
At Indiegogo, Joel works closely with entrepreneurs to guide them through the process of running a successful crowdfunding campaign and ultimate launching their new product to a global audience.
On the one hand, AI is capable of significantly improving various aspects of popular crowdfunding platforms, such as enhancing customer experience, improving data security and fraud detection (often a massive issue) and pitch assessment & improvement, so Joel. On the other hand, crowdfunding is an easily accessible and potentially highly successful tool for entrepreneurs to take the leap and bring their product to market, so Joel.
Joel subsequently provided the audience with a brief history of Indiegogo and its most successful campaigns. Launched in 2008, Indiegogo was first embraced by the independent film scene and grew rapidly to its current magnitude, incorporating every sector imaginable, from tech to fashion to agriculture.
Indiegogo itself has over the past years seen some extremely successful and also unusual campaigns. The film industry continues to embrace Indiegogo for fundraising, the film Super Troopers 2 for instance recently raised $4.6 million in funding. Indiegogo's most successful campaign of all times was a rather unexpected one Flow Hive, a 'Honey on Tap' campaign, which raised $13m.
Another wide-reaching campaign with a great ripple effect, so Joel, was 'An Hour of Code for Every Student'. With the goal of training 10.000 computer science teacher to provide 100m students with one hour of coding training across all grades, worldwide, this campaign attracted huge donors including Bill Gates, Google, Reid Hoffman and Microsoft and exceeded its goal of $5 m. This campaign also revealed that 9 out of 10 institutions did not offer basic computer training! Fatal, given the great rise of AI research and development, which requires younger generations to be equipped with the necessary computer skillset. Particularly the AI industry is in great need of highly skilled workers, which makes early education and skills acquision vital, as Joel emphasised.
Rory Beard, Phd, University of Oxford: ‘Don’t forget style!’
The second presentation of the night was held by Rory Beard, a PhD researcher in Machine Learning Research Group at University of Oxford’s Department of Engineering Science. At Oxford, Rory focused on the design of generative and inference models for unsupervised and semi-supervised learning, taking inspiration from Deep Learning, Bayesian, Information-theoretic, and Statistical Thermodynamic perspectives.
In his talk on 'Machine Learning for Separating Style and Content' Rory highlighted the notional distinction of style and content and how developers are always concerned with achieving the right 'style' for their AI.
As Rory illustrated, the epitome of unique style is to be found in artwork and painting.
Photorealistic style transfer programmes are now able to extract content from one image and style (for instance 'impressionism') from another and merge the two in a new 'composition'.
Source: Gatys et al., 2015
According to Rory, AI machine learning works in a very similar way by extracting content from training data given to an algorithm.
Many startups are currently also focusing on so called Chat bots, dialogue system which mimick the conversation with people using AI to offer a quasi-conversation for the user. However, AI still has far to go! At present, so Rory, robots can still get stuck in a communication loop, whereby the artificially 'intelligent' responses stop making sense.
Source: Deep Reinforcement Learning for
Dialogue Generation, Li et al., 2017
Another challenge AI seeks to overcome, according to Rory, is the human in-loop supervision currently still necessary during training the AI. In an ideal world, so Rory, AI should be able to replay the 'game' billions of times without the need for human feedback and supervision, which at present slows the learning process of AI down significantly.
'There are many ways to deliver the same message. The real tasks is getting these immensely complicated neuronets, which are like black boxes to do the thing we want. If we want to change an element of the sentence, how can we make sure we only change that element and not completely deform the structure? Controlled change is the challenge’ (Rory Beard)
To highlight the importance of style, Rory concluded his talk with a very amusing example of what happens when this vital element is forgotten. Watch this humanoid follow the instruction of 'moving forward' (=content), yet not considering 'style':
Q&A and Networking
The evening concluded with a very livey discussion and all attendees were left with great food for thought:
‘That was absolutely fascinating, thank you! How far do you think we might be away from translating style from for example British English to Californian English?’
‘When it comes to areas of style, where is the balance between letting an AI deal with unforeseen circumstances and being able to curate ethics?'
‘For some users they’ll want experiences which might be going against the style of the company and might seem almost insulting to the individual user. For example one passenger might enjoy driving very fast, whilst this frightens the other. Have you looked into how AI can respond to this?'
'I have spoken to a lot of people in the tech AI deep learning industry and one big question that no one is able to answer is how AI is being monitored for security?'
And next time...a very special guest awaits....!
For Meet AI series 5 on October 12th we will be joined by a very special guest… robot Olly!
Coined the 'first robot with a personality', Olly was launched on Indiegogo this week and will join us for a discussion! Olly has been developed by an expert team of leading AI researchers, neuroscientists and psychological experts and has only been shown in exclusive live demos so far, so this promises to be VERY exciting!
Our second speaker for Meet AI Series 5 is Machine Learning Engineer Yi Wei, (PhD ETH Zurich & Microsoft Research Cambridge) who will share the building of educational robots using machine learning techniques. As Yi Wei enjoys building robots and models for his 3 year old daughter, he has a great understanding of what makes AI fun, safe, affordable and encourages interaction for young children.
This is a session you do not want to miss! Find out more and book your FREE ticket in advance here.
Emotech is responsible for developing Olly, the first robot with personality. Olly is presently the most awarded robot in CES history and has just this week officially launched for pre-orders on Indiegogo and reached 100% of its goal on the first day! Not a mean achievement. Grounded in human brain-inspired AI and deep learning algorithms, Olly gradually gets to know its user and adapts its personality over time to become the perfect match. Features include productivity assistance through proactive suggestions, syncing to all home devices to streamline agendas and emotional interaction. Emotech was named a top 14 EU startup by TechCrunch in 2015.