Rise of AI Conference 2018 · Berlin

AI – a double-edged promise

Christian T. Schäfer · Deloitte Analytics Institute

At Rise of AI 2018 in Berlin, Christian T. Schäfer, head of the Deloitte Analytics Institute, argues that AI is a double-edged promise: enormous business value alongside real risks around trust, liability, security and control. He frames AI as a wide array of technologies rather than one thing, distinguishes today's narrow AI from a hoped-for general AI, and traces a hardware-driven history from the web and cloud to falling storage costs and specialized chips like Google's TPUs. He lays out risks including autonomous weapons, an AI arms race, loss of control over robust self-preserving systems, and data privacy, alongside potential upside such as recursive self-improvement, help with war, disease and poverty, and automation combined with universal basic income. He explains deep learning as a subset of machine learning that handles messy real-world data well, cites Deloitte's client demand data across five use-case areas, and closes by urging that a human-centered front end and process design, not just algorithms, determine business outcomes.

Key takeaways

  • AI is not one technology but a wide array of methods, and today's systems are narrow AI, not the general AI researchers are pursuing.
  • Storage costs fell by a factor of about 17 million over 35 years, and Google's TPUs are roughly 30 times more performant than GPUs for machine learning.
  • Autonomous weapons and the risk of losing control over robust, self-preserving AI systems are already present risks even with narrow AI.
  • Only 0.5% of all data is ever analyzed or used, despite a projected 44 trillion gigabytes of global data production by 2020.
  • Deloitte's top five client AI use-case demands are NLP with OCR, early warning systems, knowledge management, computer vision, and production quality optimization.

Detailed notes

What AI actually is

Schäfer opens by framing the buzz around AI as distracting: there is no single definition of AI, only a wide array of technologies combined to make machines act with apparent human-like intelligence. Today's AI is narrow AI, capable of specific tasks like object recognition, facial recognition or driving a car, while researchers are working toward a general AI that could match humans across many cognitive tasks at once, something current systems cannot do.

A hardware-driven history

He describes AI's history as a roller coaster of optimism after breakthroughs and disappointment at impasses, citing the web in the 1990s and cloud computing in the 2000s as inflection points. Storage costs fell by a factor of about 17 million over 35 years, enabling much larger unstructured data sets. Specialized computing hardware also drove progress: GPUs outperform CPUs for the matrix multiplications neural networks rely on, and Google's TPUs are roughly 30 times more performant than GPUs for machine learning while being more energy efficient. The underlying algorithms, he notes, have existed for decades; what changed is the hardware and the scale of training data.

The double-edged promise: risks and upside

Schäfer lists concerns around trust, liability, security and control, including autonomous weapons as AI systems built to kill, the risk of an AI arms race, and the danger that self-preserving, resource-acquiring systems built to be robust against attack could become impossible for people to control, a risk he says is already present with narrow AI. On data privacy he stresses the need to control and verify what data is collected and to instill and negotiate values with machines. On the positive side, he points to AI already automating and augmenting business processes and running critical infrastructure like power grids, and speculates that recursive self-improvement could produce a superintelligence that helps eradicate war, disease and poverty, with automation combined with universal basic income easing the transition for human labor.

Deep learning and dirty data

He describes deep learning as a subset of machine learning, using layered models that learn data representations without hand-designed features, where depth relates to problem complexity rather than model complexity. A key strength, he says, is robustness to dirty data, badly parsed, spotty or duplicate records, which machine learning can process with far greater accuracy than traditional methods, and which scales well as data volumes grow. He cites that only 0.5% of all data is ever analyzed or used, against a projection that by 2020 each person would produce 1.7 megabytes of data per second, totaling 44 trillion gigabytes.

Making AI work in the enterprise

Drawing on Deloitte's 106-person analytics team and hundreds of client use cases, Schäfer lists the top five AI demand areas from recent client requests: natural language processing with OCR, early warning systems, knowledge management, computer vision, and production quality optimization. He describes clients moving from rule-based, judgment-driven systems toward data-centric architectures, where cataloged, well-described data becomes a usable asset, a concept he calls a data asset inventory. He closes by arguing that the human interface and human-machine collaboration design, not the algorithm alone, determine whether AI delivers real business and customer value.

Quotes

The tale of AI is kind of a roller coaster ride, its optimism with every major breakthrough and disappointment when reaching an impasse.

Christian T. Schäfer

Only 0.5% of all data is ever analyzed or used.

Christian T. Schäfer

A journey towards AI starts from the front end, from the human interface.

Christian T. Schäfer

Questions answered in this talk

What is the difference between narrow AI and general AI?

According to Christian T. Schäfer at Rise of AI 2018, narrow AI performs specific tasks like object recognition, facial recognition or driving, while general AI, still a research goal, would match human-level performance across many cognitive tasks at once.

How much of the world's data is actually used?

According to Christian T. Schäfer at Rise of AI 2018, only about 0.5% of all data is ever analyzed or used, even as global data production was projected to reach 44 trillion gigabytes by 2020.

What are the main risks of AI according to this talk?

According to Christian T. Schäfer at Rise of AI 2018, the key risks are around trust, liability, security and control, including autonomous weapons, an AI arms race, and the danger of losing control over robust, self-preserving AI systems.

Mentioned

narrow AI, general AI, TPU, GPU, deep learning, dirty data, autonomous weapons, AI arms race, superintelligence, data asset inventory, GDPR, universal basic income

Keep exploring

More from Rise of AI Conference 2018

More on AGI & Future of Intelligence

Transcript

Auto-generated transcript, not edited. · 1,925 words

Show the full transcript

hello everyone thank you very much yes I'm the head of Deloitte analytics Institute and we are here in Berlin as an accelerator to do a multidisciplinary approach towards AI big data architecture and data strategy we are a team of 106 people and different domain areas sitting here and serving our clients serving federal institutions working together with cutting-edge technology providers and as well as academia to deliver solutions to our clients now I have to click perfect so this you missed I want to make an effort to frame the buzz around AI the idea that in our era we are somehow seeing the emergence offer an intelligence in silicon and that rivals our own is very entertaining it is also exciting and frightening to the same extent and first and foremost it's very distracting in fact there is no single definition of what AI actually is in reality it's a wide array of technologies that can be brought together to enable machines like machine machine computers to act with what appears to be human-like levels of intelligence

AI today is probably known as narrow AI you've heard that before in the big stage is able to do very narrow tasks like object recognition facial recognition or driving a car researchers are prone to research further into general area which supposedly beat us in most of our cognitive unlike humans and I suffer still unable to carry out multiple complex operations at the same time now historically the tale of AI is kind of a roller coaster ride its optimism with every major break su and disappointment when reaching an impasse and substitute leaf rock movements for AI include in the 90s the world wide web happened then in the 2000s cloud came available for the masses and actually helped a lot cost of storage declined if you imagine that in the last 35 years storage costs the client by 17 million acts kind of a big deal that calls actually for larger unstructured data sets to be tiered according to the basis of data users patterns and performance requirements of applications here by mapping the storage and

data uses patterns to the business requirements also we've seen a surge of innovation when it comes to computing course so CPUs are dwarfed by the ability of GPUs however there's so much circuitry on a GPU that's actually not needed for machine learning because neural nets are quite simple they are matrix multiplications and nonlinearities that you can build silicon directly to do that Google uses TP use for some of their more Nvidia machine learning approaches in cases and a TPU with 30x more performant and it is much more energy efficient also most of you in this room know that the Abrams underlying our current state of art actually developed for decades now the only difference is that is put on a higher performing hardware and are trained in much larger data sets now designing smart AI systems in itself is a cognitive task and there is a certain ambiguity to that that's why the title is the double edge promise along the lines of trust liability security and control they are very prominent Moises warning us

to be mindful when thinking about AI to name a very very few concerns keeping in mind that this event is called the rise of AI I couldn't resist to quote Skynet in the making let's assume that the drive for self-preservation and resource acquisition is inherent in all systems of a certain degree of intelligence and with techniques like deep learning laying the groundwork for computers to learn more about the world around them how can we control and guide machines if this is the case for the ethical concerns if he is programmed to do something devastating and I'm here I want to reference autonomous weapons these are AI systems in the end program to kill in the wrong hands this is very bad news moreover an AI arms race could invert early lead to an AI war to avoid a hostile takeover such a machine these machines would be built very robust and that would make it quite hard for a person to attack and overtake on the other hand it also would pose a great risk

of people losing control over these machines the risk is even present now with narrow AI and will only substantially decrease if autonomy intelligence increases data privacy now I'm not gonna take a step and talk about next week's Friday GDP are kicking in I'm more I'm more on the point of saying that we have the ability to control and to know what data is being collected ensure that the data that is collected is correct updating information or provide the necessary context to misalign values now the question here is how do we instill values in machines and how do we negotiate with those machines if and when we believe that the values are likely to differ from our own now to the positive side of things and I could ramble on for let's say negative impact and scenarios for a long time but I don't have the time now if if such a system could potentially undergo recursive self-improvement triggering an intelligence explosion it would indeed be the ultimate search engine bidding us at our own game

driving competitive edge our clientele is largely embracing and translating I to add substantial value to automating and augmenting processes and services already and the aura around us the way we navigate consume decide operate even running critical infrastructure like power grids is largely supported by AI the journey has long begun and its prime to penetrate and scale on enterprise level absolutely deserving to be classified as an arms race with very little potent players and playing a leading role it is won by talent intrapreneurship and accepting controlled risks through the domain of service from mankind if the super intelligence would be coined it would probably help eradicate war disease and poverty and so it could be one of the biggest events we can witness from agriculture to medical diagnostics it will support and is already supporting the well-being and welfare of society creating such a search of human creativity with unprecedent new possibilities will help us reinvent ourselves and maybe become something different also there's lots of talk for the evolution of employment daddy is gonna steal

all our jobs away well I do believe that in combination with Universal pay and you found love for that human creativity we've all missed out when looking into an SI p interface it will beat the changes human laborers undergone so far by a lot now I want to talk not only about dirty data I want to talk about these concentric circles over there so machine learning and deep learning as a subset for machine learning at the core of every AI system deep learning allows the computational models composed of multiple processing layers to learn representations of data with multiple levels of abstraction so the key here is that these layer features are not designed by the human engineer but rather using the data and general-purpose learning techniques now the depth in deep learning is related to the complexity of the problem not the model one critical advantage machine knowing has is that it is very robust to derrida what's dirty data it's badly parsed spotty missing information incorrect values duplicate records every executive will know that

dealing with dirty data will be and is the bane of their professional lives machine learnings flexibility and ability to learn and improve over time means that their leader dirty data and be processed with far greater accuracy it also means that the technology scales very well something that has become very important in our era of ever increasing data volumes now a fun fact is that only 0.5% of all data is ever analyzed or used now if we put that into the relation of that by 2020 each human being on this planet will produce 1.7 megabytes of data per second resulting into a surge of 44 trillion gigabytes and this is quite big but it's not smart but there is lots of potential now to the centerpiece of this and coming also slowly to a close how to make it work is a big word I I work for a consultancy this is part of our DNA we try to figure out and help but I want to twist it in this scenario a little we take

this stance from having worked on hundreds of use cases in the past years also we take the stance from our multidisciplinary approach to tackling such a theme like AI um our clients have embarked on this journey to seeks to increase just a value at is see in the first promising steps towards AI and to give you an idea of what the vectors are that our clients are requesting here the top five of the last five four months so on the first spot is NLP with advances in OCR the second bit is already warning systems the third bit is knowledge management and then there's computer vision in ODS applications and production quality optimization so this is the demand side of our business and with that we can we can clearly say that clients are moving slowly away from expert judgment systems largely embracing rule-based systems supported by RPI and robotics and turning towards a windy and very very cumbersome past or which judgment based a solution adhering to the convergence trend and big data architecture data

is put in the center of gravity now data centricity actually calls for a systematic collection and catalog logging of data assets now they can only become an asset if they are ready to be used to generate business value I talked about dirty data before and you all know these haystacks where you try to find a needle in first you build the haystacks or you define the needle I'm more into the picture of a warehouse where you can clearly describe and see what you find in each you know compartment it's like bread sugar and milk because you need to understand the potential know how to read the data and only to harvest that and put that into value add with the help of that meaningful metadata in a central catalog data can be found understood and used even beyond its collection and that's the most important part is last words of that sentence I try to sketch it here away from the haystack to a data asset inventory coincidentally it also reads dai which stands for

the Deloitte analytics Institute let's call this the data as inventory for now and let it be flanked by a comprehensive strategy and operating rhythm to ensure that your data doesn't become your liability and on this note I will present the last of my slides and I want to close with the most important aspect to me personally when bringing such technology likeéi to life a journey towards a nice starts from the front end from the human interface here by taking well measured strides towards business value and customer value the human-machine collaboration process architecture and design make all the difference between the output of an algorithm and the actual desired outcome so in this sense think of Margaret and think of John

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.