This is a quick canter through things I learned cognition X, conference and celebration in London dedicated to technological solutions to modern problems with a focus on artificial intelligence
There is not much structure to this, as I made the notes as I went from event to event, seeking serendipity. Nor do I claim these are stenographer level accurate records of what was said. You can, of course, look up what was said in the cog X videos. This is more my interpretation the key themes than transcription. I have, however, attempted to bring it all together into something that I think indicates where things are going. Perhaps you will agree with me, perhaps not.
What do I deduce after CogX?
“The future is already here — it’s just not very evenly distributed.”
- William Gibson
There is no doubt from the evidence provided across a wide range of experts that Western civilisation is at a key moment of change. It is no longer in control of its physical environment, politics, social media, or consumption. Without several whole society level changes operating in parallel, it is doubtful whether the current global civilisation will survive climate change, a radical and complete change in technology and the world of work, bad actors undermining morale with fake news, the collapse of once stable and reliable political structures.
Very few people truly understand exponential change. Most particularly, they don’t understand that exponential change is not only in terms of scale (numbers of users or events), but in terms of speed (things happen faster and faster). I have personal examples of large banking and insurance companies that have fixed and rigid decision-making processes for new products and new operating processes that require a stop to deployment time of around three years. Technology is currently throwing out revolutionary new opportunities every 18 months. That will soon be 12 months. Then six months. It is now so fast that those who adopt new technologies later are liable to overtake those who adopted earlier technologies soon. And this is creating absolute paralysis in politics, business, and daily life where the huge range of choices and the massive consequences of making the wrong choice overwhelm the systems we have built currently.
Artificial intelligence – by which we only really mean some forms of machine learning, pattern matching, and prediction – has come along at the perfect time to be our tool to get ahead of exponential change. It is one of the first tools that humans have built that can improve itself. Currently, it is horribly energy inefficient, deeply dependent on the quality of training data, opaque, and requires a priesthood of experts to build and maintain it. This is likely to change extremely fast. And when it does, artificial intelligence will penetrate every aspect of business and life extremely rapidly.
The good news is that if we can harness it, it might help us provide the problems to global energy crisis, global climate change, food distribution, politics, and business planning. Sadly, standing on its way are appallingly old-fashioned measurements of “profit” that reward destruction of the ecosystem, exploitation of society, and the dumping of pollution. Until we fix the way we measure things, venture capital and business will not be able to allocate resources to save humanity and create a sustainable future.
The sustainable development goals provide an excellent framework, albeit they might be worded slightly better, against which any Board of Directors, any private equity fund, any venture capital firm, any ultra-high net worth individual, and any concerned citizen could measure their behaviour in every aspect of life. If everybody did this, we might all make a small positive change, and though small positive changes might give the large positive changes time to act.
Venture capital needs to work much more closely with government to ensure that capital is allocated to research that will yield globally sustainable positive changes. If it goes on simply trying to extract profit from the next social selling app, then it is doomed as well. Currently, almost all private capital is built on government expenditure, infrastructure, and research. It should pay back.
AI and software alone will not solve all our problems, and in their current iterations rather poorly designed and inefficient. We need to reach interrogator will artificial intelligence – something that can explain why it decided something in words we can understand – and we need to make the energy cost of running machine learning two or three orders of magnitude lower.
We do not have a choice about this – the decisions have already been made and the forces are already in unstoppable motion. The only question is how we adapt.
If we line everything up, and work together then – and only then – there is a small chance that we might transition to a really positive and sustainable future; one filled with music and art and creativity and human meaning.
the United Nations 18 sustainable development goals provided a strong opening to the event and that theme was clearly continued in each of the dozens of minor stages, demonstration areas, and partner expositions de Khan, Mayor of London, gave a balanced introduction around his hopes for technological solutions to issues of growth, quality of life, society’s safety, and the deep and profound issues that artificial intelligence raises in the world of work he exhorted us to go on a mission to augment and enable human beings, not to replace and control them.
- Aristotle predicted that machines would replace men something he wrote 20 BC
- Babbage suggested in 1842 that machines could learn, and would one day replace mankind
- Thornton in 1847, and Turing in 1951, fully expected intelligent machines to replace men
- finally, in 1956, human beings started to study what artificial intelligence was and how it might be made
AI only becomes useful when it can guess human objectives, and then do what is reasonably necessary within a series of constraints to achieve those objectives. A machine which has to be erected, de facto, not artificial intelligence. A machine which merely answers questions posed to it is not particularly intelligent but may be artificial.
All our current artificial intelligence seems to be based upon for historic themes of study:
- control theory – in which we seek to minimise costs
- operational research – in which we seem to maximise rewards
- statistics – which we minimise the secluded parts from our functions formally
- economics – in which we maximise a utility function (usefulness)
Each of these is partial, flawed, and brings with it a host of biases and inconsistencies. The leaders in the field hope that we can move artificial intelligence from statistics towards probabilistic measures supervised learning into reinforcement learning and then into probabilistic programming (and this appears to be the stage we are just entering)
we are a short generation away from intelligent systems that have access to:
- Web scale knowledge (that is everything currently known), and
- the global object database in real time (knowing where everything is from satellite data down to the level of an individual sheep)
Despite that, we have major missing parts in the chain to any sort of general artificial intelligence or any form of intelligence at all at the human scale of reasoning. These problems show no current signs of falling to the tools we have today. Because of that, there is a great deal of uncertainty as to when we may have any form of artificial intelligence beyond the current tools which – frankly – merely match things to patterns and describe names to them
it is unlikely that we will have general artificial intelligence within 25 years, might have it in 50 years, and we probably will have it in around a hundred years. But we don’t quite know what we must do to get there.
Why would we bother getting there?
When we have generalised artificial intelligences, all the prediction show that global gross domestic product will increase by around factor of 10 X. Artificial intelligence is worth at least $13 trillion per year.
[That estimate, sadly, assumes that we keep it out of the hands of bad actors, lazy actors, civilisation ending rogue artificial intelligences, and dictators.]
As little glimpses into what we can do already:
- real-time global forest monitoring, so that we know exactly how many trees there are on the planet all the time and what is happening to them
- reducing clinical errors by 90% saving millions of lives around the world
- augment and adapt education at the level of an individual to greatly improve their lifetime outcomes and learned skills
Kate Raworth – author of “donut economics”
tied into the UN SDG’s – she pointed out that artificial intelligence can help us deal with the nine limits to growth. All the major ecosystem issues are amenable to human repair, having been caused by human action. What to do, and could do it cheaply enough, and we could solve the problems quickly. Surely, therefore, humanity should demand that we use artificial intelligence to solve these problems?
We must end the “take – make –use – lose” cycle of exploitative economics. We need to use artificial intelligence to create regenerative designs, more efficient recycling, better resources, and more efficient logistics.
Our current economy is cumulative. It should be distributive. Perhaps we can make our economy “reuse – make – reuse?” There is a massive design opportunity for anybody who wishes to help get this right.
Books that can help you get your head around this
- donut economics by Kate Raworth
- the entrepreneurial state by Mariana Mazzucato
- the value of everything by Mariana Mazzucato
Organisational capacity in government research is a critical resource for any technological revolution. The hard truth is that the private sector never actually invented anything, but they accelerated it into commercial use almost all new products and discoveries, new technologies, new engineering, and new compounds start life in government funded research in a major nation stop
State funding for innovation currently runs at 3 to 5 times larger than the total worldwide venture capital investment innovation. The problem we face is partly that 20% of VC money has started to lobby for the 80% of government money and innovation to fund things so that it can make profits from them. This is heavily misguided. Government innovation should be targeted at social good, sustainability, and regeneration – it should not be just another resource for capitalism to extract value from.
Tax incentivised investments just improve the returns to investors, while simultaneously miss allocating scarce financial and talent resources to the wrong problems. The world simply does not need yet another way for advertisers to reach board customers with pretty pictures. It needs solutions to existential problems
Whatever we do, we’re going to find that there is no one solution – we need a system design approach to the SDG’s as they are too big for anyone answer
Unconstrained digital technology has transformed the state, but fundamentally undermined the state. It has eroded the power of the state and the ability of democracy to represent the population it has reduced economic friction to near zero and allowed unconstrained accumulation of financial assets post 2008 the Western world has chosen to enter an austerity trap which has been partly enabled by superefficient digital transfer of wealth from the poorest to the richest.
By 2016 we thought the US had seen what to do, but we are now in the middle of losing 4 to 6 years to the disastrous policies of Pres Trump.
We need the public to stand up and demand a mission led politics for the future
The truth is – no one is in charge. Must coordinate our own actions, and urgently. There is no good answer: we must learn by doing things together. There is no reality that permits any other path to succeed in the time that we have available after all, the Paris climate agreement took 23 years of work, we’ve only just published the SDG’s.
Take these fragile, incomplete, missions and gently try to do them together and improve them as we go along.
It is this simple: humanity lives or dies by these simple tests. So perhaps AI can help us to simplify complex analytics? Can it tell us where to put our money to save humanity? Can it help us to allocate talent, resources, and logistics?
Accounting and economics actively prevent us from doing the right thing. Sadly. So, resources are not being mobilised to solve the problems that really matter. Governments, in turn, are obsessed with capital accumulation and libertarianism. This actively prevents them from spending money in areas which would otherwise benefit humans through the SDG’s. Ultimately, as ever, the problem is politics.
“Brexit is the most tragic and absurd diversion of human thought in human history”
modern capitalism started in 1587 with the creation of the nation state and the invention of piracy, whereby the corporate venture for profit was unleashed upon the world. Even upon that time Sir Francis Drake was not responsible for his losses – the state had underwritten them. Libertarianism has not changed its method since the days of piracy.
The 18 grand challenges in the SDG’s demand not just technology huge social well and coordinated political action.
Against this background, tax evasion is a global catastrophe. The UK is about as corrupt as it gets by way of money laundering, graft, and secretive lobbying. None of these issues can be solved by any one country, and countries that wish to solve them need to form larger alliances.
Now is the time for cooperative action across nations.
we have the roadmaps, but who will lead the world along the roads who will help build the new roads of technology that will take us into a sustainable future? We properly have all the tools we need, but lacks the capability to deploy them – or – perhaps we are simply uncertain as to the direction we wish to deploy the tools we have?
Representative democracy is young, and very fragile. Artificial intelligence could perhaps help us to fix this. In China has become a tool of tyranny. Can we stop that happening here? Will guide the development of artificial intelligence alongside democratic processes?
Doughnut economics is demanded by civil servants and by most of the people with any informed decision on the future of the planet but 1% of the world’s richest sociopath resisted fiercely choosing, instead, to accumulate assets as some sort of peer group scoring mechanism. No matter what they want, we are the majority and we will get there. The question is what are our priorities and how are we able to set them against the evidence when so many people work against us with such large pools of resource?
The changing nature of investment
- Kim Polese from Crowdsmart
- Ali Tamaseb from DCVC
- Nic Brisbourne from Forward Partners
Better VC decisions are made when they introduce cognitive diversity into the pre-diligence stage. This allows for quicker pattern matching. It also runs a risk of group think and availability bias plus a touch of bandwagon chasing if not controlled. AI pattern matching and predictive ability is a useful adjunct to this and needs access to machine learning and big data on investment outcomes and company behaviours pre and post investment decisions.
The sad fact is that VCs become less evidence based and more biased as they get older.
Primary AI and similar companies continue to develop new AI paradigms and methodologies, and to integrate old ones into new tools. Things move fast, and investment into any one tech is risky for VCs.
Real value exists in tracking investment outcomes against predictions at moment of choice. AI has become a pre-requisite for being a successful VC, part of the formula, as money and people and software was until 2015.
[I kept thinking: “However, this still delivers all the capital and intellect and effort to the “most profitable” opportunity where profit is defined by a broken economic and accounting model. Worse, legal structures actively subvert sustainability and encourage owners to accumulate capital while exploiting and extracting value from natural and human resources without making provision for clean up or remediation.”]
VCs ride on government investment and University ecosystems are a critical part of the marketplace. While many VCs and PFOs want to associate with Universities, they are linked to their ‘home area’ and find it hard to commit to being physically co-located with Universities, and this limits their insight and effectiveness.
Despite a bump up in early stage money, the ‘valley of death’ has not shifted to Post A round, and part of the problem is too much early capital is keeping too many bad ideas and bad teams afloat for too long. In those bad cases, a common cause of death is the valuation trap which then pushes a ‘down round’ and kills all motivation for the Founders and team.
As VCs have become more aware of risks, Due Diligence has become the rate limiting step for investors. Hence the drive in the market to Machine Learning scores and cognitively diverse assessment in pre-diligence.
Investors can positively influence the outcome by providing a platform of services:
- Services (back office)
- Access to market
- Exit planning
Combining the two (MK and service platform) seems to increase returns and reduce losses.
Health and AI
Rich people are getting richer and healthier; poor people getting poorer and sicker. This is not sustainable. It is partly driven by the huge fixed cost of medical diagnostics and imaging.
[I’m going to caveat this one with a suggestion that the tech was obviously there in ‘sales mode’ intent on raising major finance. It had a little bit of a ‘Theranos / Woo” feeling about it, where the visible results were not on par with the theoretical claims. However, all it the benefit of the doubt for now].
What if we could image brains and bodies in real time and connect our brains to machines to diagnose our state of health at a profound level? What if we could personally image inside our bodies with a handheld device in real time, and connect those images to powerful AI?
The proposition is focused holographic red-light and IR laser + Ultrasound, from a handheld device, and aiming to create a drop in the price of imaging inside the body of 1000x over 10 years ($1000 to $1 per image). They have 25 patents in hand already and are moving forwards to collect data to train a recognition AI.
This can be external + internal imaging: trials already running to help wound nurses assess abscesses in elderly patients for healing and match that image data to the range of motion and limb use from accelerometer data. This enables patients to go home faster and heal faster. There is already and ML phone app that can “listen” for the distinctive sounds of Tuberculosis coughs.
This is against a background in which they expect and prefer that AI is used to inform and de-risk the decisions of qualified medical staff. Not replace them. We could aim for AI to have us find better whole life HEALTH plans and reduce the total lifetime cost of health intervention. One simple way is to keep people compliant with treatment and lifestyle plans. NLP + mobile devices + machine learning can create a ‘doctor on your phone’ … it just needs sensors to complete a remote diagnosis.
Early intervention is the number 1 key to reducing the cost of health management and is hugely socially valuable. AI has to be part of that.
Looping back, we have to be careful NOT to create wealth traps for health by only focusing on high value solutions (diseases of wealthy old people that need expensive drugs to treat.) Yet we know that if everyone gets to live to 120 years old, then we will need a lot more complex, interactive, intrusive, sensor heavy treatments to keep us all alive, aware, and active. AI and ML will be part of that, but when it is, how do we allocate the costs of creating, building, and training those in a fair way to the benefits released? Will be we tempted to segregate our population into treatment groups based on objective data?
If we have genetic data, we can show that gene expression varies by genetic legacy and population factors. ML can see through the clutter into the underlying structure of the information.
[Sounds a lot like GATTACA to me]
These new tools have very fast take-up: Samsung got 80% take up of Laparoscopy in just 5 years.
As a side bar, this business will rely on a sustainable delivery model to customers, which in turn probably requires AI to define the logics and operate scheduling in real time.
- Salim Ishmail
People find it hard to understand the exponential nation is succeeded by another technology each time the “S curve” of adoption flattens out. It is a series of ever more rapid changes stacked upon each other.
This difficulty in understanding as part of the mechanism that ensures that any changes fiercely resisted by the “corporate immune system”. However, if we can harness this expectation, we could live in an age of abundance within 10 years.
Approximately 12 technologies are now on the standard doubling curve, each of them has the potential to radically transform human life. The things that drive this along are disruption (powered by digitisation, democratisation, demonetisation, and distribution). If we can harness these technologies we can move from scarcity to abundance, from costly to free, from governed by two governing, from expensive to affordable, from physically constrained to digitally free. In each time of change, what we are doing is moving the problems that we had into a new domain. (We solve one problem, but we reveal that there are other, deeper problems).
Solar energy will be able to provide 100% of global energy at the lowest price ever experienced by humanity in just about 12 years. That means Russia, the UAE, Saudi and the USA are likely to suffer a serious collapse of their oil-based economies within 25 years. It also means that within 50 years, energy is likely to be free at the point of consumption (other than for distribution costs) highly abundant, and mostly generated in the poorest countries which also happened to be the sunniest countries.
So, what about investments in the power distribution utilities and large nuclear power stations today? Does that make any sense? Are there better ways of bridging the current state to the future state that are more efficient, more effective, and less polluting?
Whatever we decide, software will not stop.
Institutions cannot function in this world of the future. Everything we know about banks, churches, and Western governments will fall apart under that amount of disruption. That is, unless, they choose to act now to embrace, understand, and adopt it. They will not survive if they do not change.
What then will happen to our pensions when all stock markets collapse around the world?
For some good news, robots and artificial intelligence have not ever reduced employment. There is always a knee-jerk reaction to ban it from the regressive right-wing reactionary elements, as well as from the ultra-left-wing elements. They have a Luddite view of new technologies. Equally well, the richest 1% do not want to lose their wealth and power but have not yet transitioned to ownership of artificial intelligence and robots. That transition will be very delicate and dangerous for them.
Sadly, the reality is, corporate executives punish those who say these things. Every.Single.Time.
This corporate blindness is preventing humans from moving forwards, anchoring us to outdated and inefficient, polluting, undemocratic, physical thinking and archetypes.
No matter how hard they try to resist it, the future is somewhere along the exponential curve and one of open everything.
Global investment trends
- Ibrahim Ajami – Mubadala Ventures (now has its own venture arm, £200m, split across EMEA with focus on tech)
- Dharmash Mistry – private investor
- Christian Angermeyer – libertarian director of Aperion Investment Group, Brexitard pro-Trump fratboy (IMHO)
- Louise Tabbiner – Introsight, deal source pipeline for VCs
- Julia Hawkins – Local Globe
Europe is vastly more capital efficient, although US markets are much larger and have lower capital friction. Europe therefore looks cheap to venture capital, but it’s very high growth is attracting significant new money which is having a price distortion effect. Europe now has its own second and third generation scaling teams of executives who can provide real value creation in their next companies, as well as providing support to smaller ventures from their own investments.
85% of UK investments at the sea level and tend towards the London area. The UK has created roughly 34% of the VC unicorns and the European Union. Despite being only 10% of the economy.
UK founders tend to sell far too early and then buy a house in the country. They should go on and found her second and third companies. EU founders avoid becoming large enough to attract state attention, and there is very much a community-based investment approach in most of Europe.
There appears to be a cultural disjunct between the sustainable development goals and the ambition venture capital, and some venture capitalists just want to accumulate personal fortunes, billions of dollars and political power.
[I really don’t think this is a good thing myself, but, hey, some VCs do!]
The next 10 years will be rough, but clever players can grow and thrive. Government should focus on local gross value-added, local ownership, local leverage, local growth patterns. This way they can balance inward investment with foreign talent to benefit their communities. Governments need to listen to and involve early-stage CEOs more in their environmental and social planning.
Government should enable and reduce friction as much as possible in marketplaces. British business bank has turned out to be a relatively good idea.
A lot more thought needs to go into the long time it takes to move an idea from University thinking through research and development to a prototype and then into a customer market match. Only at the point that the customers met the market can VCs step in and fund the business into growth.
Right across the European Union long-term R&D lead times are long, and pure research and R&D need lots more support.
[I think this is a plea for more hosts to parasitise by venture capital myself. Why does venture capital never fund research?]
Building moon shots
Julie Hannah and Astro Teller of X factory/moonshot
Diversity is critical to success.
Ethics need to be baked into the team
Test in the real world early
Listen to your customers
Building billion-dollar companies and genuine moon shots is more driven by a sense of mission then it is by any sense of future financial reward.
Competitions can provide those missions.
manufacturing can return to city centres via modern manufacturing technology, maker technology, and 3D printing. Farms can also return using vertical farms LED lighting and solar power. Drones allow us access to the third dimension of cities both in time and space. They allow data gathering on demand and at scale. They work best when combined with static sensors and satellites.
All these things require mega-scale connectivity and bandwidth. That requires government a major institutional capital.
As this deploys, we will see major physical changes to our 400-year-old built environment across our cities. Most of the things that we regard as “new” in our cities are between 25 and 100 years old.
None of these changes are sustainable without the political support of the people who live there, and the citizens will be asked to accept massive personal change, change will accelerate, economic disruption that will threaten things that they think are important because they cannot yet see the benefit of the future.
One potential benefit might be a shift from worktime to community time? We may not be employed by a company, but are we going to be paid for helping communities?
Buildings currently take between 10 to 30 years to plan and build inner city area. Developing them using current tech is extraordinarily slow, high risk, and has generally poor outcomes in terms of financial planning community involvement and aesthetics. Compare this to the lifetime of a technology (currently 3 to 5 years). How, then, are we to build smart buildings if the technology that makes them smart is already 3 to 5 generations out of date when the building is complete? How can we build buildings fast?
Whatever we do with the built environment, we should focus on the public realm, shared spaces, and aesthetic set step one in planning. This way, we can bring people with us and will harness their imagination and creativity to help solve the other problems.
There will be a major drive towards another sort of city “super green, super low energy, superhuman friendly”. The cities will be built on the outskirts of existing cities or in green areas.
The climate crisis
Kate from extinction rebellion
Director from Reckitt Benckiser
[both have been brilliantly covered in social media and both have extensive PR and advertising outreach, so I don’t think I can add anything by describing them here other than to say the issues are obviously important, urgent, and existential. Moreover, the solutions to the problems are not obvious: it may make sense to have dishwashers in sub-Saharan Africa or to stop an airport for a day in pursuit of larger longer gains.]
Technology to save the world
We should all go away and read Edgar Bronfman Jr – “the global thermostat, will capitalism save the planet?” – Fundamentally the hypothesis is that no matter what we do, or how fast we act, we must find ways to sequester carbon out of the atmosphere rapidly because we are already over the tipping point for climate change. Reforestation, gland, and high persistence crops may have part of the answer, but the other part of the answer almost certainly requires mega-scale engineering.
Is it possible for humanity to sequester 30 billion tonnes of carbon dioxide a year at a cost that makes the carbon useful rather than a burden? What would we then deal with the billions of tons of carbonates when we have them?
If so, who is going to put up the billions of dollars necessary to make this happen?
How big data and art collide
Technology might save us through art – as we gather more spectral and physical sensors and apply them to artworks we can end up with large data cubes describing every layer of material, its age and chemical and physical compositions, the direction of every brushstroke, everything that happened to it since it was created. This is a huge data cube across every element of spectral analysis from terahertz radio to x-ray.
We will then know more about art than our eyes can see, or our senses detect. Will this change the way we understand art, or the way that we make art, or not?
Could we then feed this information into a generative AI that would make art? Would that be art? Would humans respond to it emotionally?
Artists are currently feeding their work to adversarial and generative artificial intelligence as datasets and then recursively using the output to generate new things. The final output is still human art, but they have been informed by the AI. Should we regard artificial intelligence is anything more than a simple tool like a brush? These methods work not just for paintings, but for music, video, photography, and writing.
The problem is that all the current tools reduce the art to simpler, more constrained, more abstract things. But this might not always be the case! What if art created by artificial intelligences was more human than that made by humans?
How artificial intelligence and science collided
Artificial intelligence has begun to work with multiple data types: networks, timeseries, images, sequences, numbers.
Artificial intelligence researchers are looking at new ways of training machine learning and artificial intelligence that are more energy efficient, more data efficient, more replicable, and more understandable. We cannot yet fully interrogate and artificial intelligence to understand how and why it decided, but we need to get there if we are to rely on its outputs for science, governance, or commerce.
Data science plus artificial intelligence can help us to plan and adapt to climate change. We can modify the environment of plants to convince them that the future weather will be just great and then they will grow optimally.
We can feed information about genetics and the environment into plant sensors against a growth model of the internal state of the plant and predict what will happen to a crop. For instance, working with spring onions, we can see that they remember the temperatures they were exposed to when small.
Taking all these things together, we are rapidly gaining the tools to be able to modify our environment not just for crops, but at the mega-scale across large natural plant ecosystems.