The doctor is ready to see you…tube videos

How many lives could you save if you stopped wasting your ML talent?

“The best minds of my generation are thinking about how to make people click ads” — Jeff Hammerbacher, early employee of Facebook.

“Biology is incredibly complex — maybe even beyond the ability of the human mind to fully comprehend. AI-powered platforms have the potential to connect dots that have before looked like noise; to generate new discoveries; even to change the nature of discovery itself” — Jorge Conde, Vijay Pande, and Julie Yoo; Andreesen Horowitz, leading Silicon Valley VC firm.

“… we tend to forget that Icarus was also warned not to fly too low, because seawater would ruin the lift in his wings. Flying too low is even more dangerous than flying too high, because it feels deceptively safe.” — Seth Godin, oddball in the greatest sense of the word.

Knowing that two hands are a lot, I’ve decided to use both of mine typing out what I think about on a random Tuesday below (we’re hiring).

What’s the meaning?

Is what you do meaningful? No, really. Suspend your society-normed autopilot reaction and think about it. Not “do you earn a lot of money”. Not “is the firm or institution you work for prestigious”. Not “does that place talk ad nauseum about ‘impact’ trying to convince you they have some”. Not “does that place have an expensive interior and great lunch”… But: is it really meaningful? The stuff your grandchildren will tell their friends in kindergarten legends about. Let’s be honest: they won’t say “My grandma is the best… she increased operational efficiency of [other meaningless company] short term, through restructuring their innovation department.”

An Austrian nurse spent years recording what people’s biggest regrets were on their deathbed. One regret that came up again and again — spending too much of one’s life working and doing things that were not meaningful.

Surely these people, had you asked them in the middle of their working life, would say that things were not too bad; that they had many of the ‘badges of society’ — but worth nothing in later life.

We have a meaning problem today. A big one (even ML might not be able to solve it). The most highly skilled, extraordinary people of our generation are wasting their time in profound ways.

• The best research engineers, machine and deep learning researchers and software engineers are being enrolled in the armies of the world’s biggest superpowers — Facebook, Google, Apple, Amazon, Baidu, Alibaba, etc — in the arms race in the war of attention. The world’s best engineers and ML scientists are working on turning mobile phones and computers into dopamine slot machines by optimizing intermittent variable rewards. Working on optimizing ad-clicking by exploiting private information from unsuspecting users. Working on supercharging binge-watching algorithms to help Netflix fight against their alleged biggest competitor — “sleep” (Reed Hastings, Netflix CEO). Or anything else that is at its heart malicious, or even just banal — and thus a waste of the great people’s time. Why do many of our generation’s smartest people spend their lives or best years turning these companies with fundamentally questionable impact into the biggest monopolies ever to exist? I mean, the lunch might be good, but…
• Some of the best doctors and scientists are getting lured into management consulting or investment banking (etc.) through designer-chair-shaped sirens and a voluminous compensation packages; where they proceed to have their creativity, scientific curiosity, and personal missions chopped to pieces with excel sheets. A Physics PhD now researching how the oil industry can fend off and crush competition from renewable sources. A doctor diagnosing what part of the body in an organisation that can be amputated.

One thing that exacerbates the impact of this problem is something known as the Pareto distribution: the square root of any given amount of people will create >50% of the value/output of that group. In 10,000 people the ‘best’ 100 will create >50% of the value/output. And when these 100 people are occupied by making ever incrementally more addictive ads, society is missing out on a lot of value. Surely there are better places where your scarce skills could be used.

Come on, ‘best 100’ people — we need you working elsewhere! (and no, not being bossed around and binned on Downing Street)

How about a real problem? A hard one.

We’ve got a real problem for you to work on. You, your family, people of the world are getting older. And they’re getting ill. And there’s not much being done about it today.

Do you have a parent or grandparent with dementia? Have you experienced how it’s robbed their memories, what it means to be them? The impact it’s had on them and the rest of their family? Have you known someone depressed or anxious? Or suicidal from bipolar disorder? A family member who can’t eat properly because of the tremors from their Parkinson’s?

We have. One version of a much too common story: my good friend and co-founder Jack Weston lived through his childhood with the experience of his grandmother, a native Finn, slowly being robbed away by the disease. As the disease ate her brain, she gradually unlearned first her English, and then her Finnish, until she wasn’t able to communicate at all.

What makes going through this infinitely frustrating is that there’s not much to do about it. For instance: Alzheimer’s disease and other dementias will affect 1 out 3 people (~same as all cancers combined), yet despite decades of research and billions spent on drug development, there isn’t a single disease-modifying drug on the market.

And what makes this situation unbearable: we know why there’s not much to do about it once you know you have the disease. Alzheimer’s is diagnosed much later (decades) than it begins. We know this from clever studies looking at subtle changes that occur in people many years before classic symptoms appear.

One such change is in the way people speak. A study analysed old diaries from Nuns that developed Alzheimer’s later in life. Here, changes in written language were present multiple decades before the disease became clear! From this and many other studies we know that there are early changes in the way people speak, that are characteristic not only for Alzheimer’s — but also Parkinson’s, MS, ALS, Depression, Anxiety, Autism, and many more. A human has a hard time detecting these changes early. But machines can.

Now next you might be surprised to learn that many tests that are used today to tell if someone has a brain disease were made up in the 1900s. We actually know that they don’t work well (this is not any secret, but something most doctors will sigh when talking about). And especially not at the time where these diseases have a hope of being treated. In what areas would using 50-year-old technology be broadly accepted? Here we’re dealing with the health of your loved ones.

There’s many brilliant people in medicine. And many that want this to change. Over the past year we’ve had the pleasure of working with many of them, both from leading pharmaceutical companies, healthcare systems and academic institutions. These industries are changing, taking a more open-minded approach to collecting data on patients, particularly from digital sources. Some are ready to change a lot. Where a few years ago you’d run into a brick wall, today that wall is being torn down from the inside.

But there’s a need for some particularly brilliant, unconventional, world-conscious, collaborative, exceptional individuals to help make sure this translates into something that will help patients. A big part of this is making sense of these new data streams. And doing so in a human way. This is hard. We hope you’re one of these people.

The human rewards here are clear. One has to only look back at the miracle developments we’ve seen in cancer drugs, after next-generation sequencing pushed us from thinking about cancers as defined by organs, to cancers as defined by what’s actually wrong with them (their genetic signature). Objective, digital profiling has the promise of causing a similar paradigm shift for brain diseases and conditions.

There’s a lot of BS ‘solutions’ out there. Many ‘digital health’ startups, self-prophesized experts and consultants, and other players selling digital snake-oil. This is one of the reasons we’ve not seen much real progress. To be clear: if you “are looking for an interesting side project”, “want to ‘break things’”, “see no problem in the way big tech companies deal with personal data”; if you think “it would be great to passively collect data on people through a ‘digital health app’ that is not illegal only because “you’re not making medical claims”, even though that’s what the users think, and that “there’s absolutely no need for doctors (or maybe even the FDA/EMA) now that we have ‘AI’” — in short: if you’re looking for an easy way to ‘solve something’ in this area, because it’s “hot”; if you care more about the title of the ‘project’ on your CV, than the people it will help; then this is not for you.

On the other hand, if you’re looking for a fundamentally hard problem to solve. That others have not been able to solve before. Where even at the limit of what you’re capable of, it might still evade you. If you want to do this in the proper way. Lay the groundwork and build the foundational work that we know are needed to make something that will outlast you; if you’re more scared of flying too close to the water than to the sun; then this is for you.

What we do is simple on the surface: we use the way people speak to assess whether they have particular brain diseases, how well they’re doing, and how we can treat them. Beneath the surface there’s some incredible hard, technical, and human problems to solve to make this work. To make it really work.

Why speech and language? Well, people speak all the time. It’s the core medium of information transfer between human beings, and this includes when you speak with your doctor, do a neurological assessment, and when you interact with your digital devices. Speech and language are innate abilities, fundamental to being a human; evolution has dedicated much cognitive real estate to these functions, that involve coordination of multiple domains. For these reasons it’s turned out that speech and language has the potential to give the best system level readout of the state and health of the brain. And this, importantly, holds true across dozens of highly prevalent brain diseases. Finally, technology is just maturing to the stage, where we can properly listen for these changes in speech and language that can occur decades before formal diagnosis.

We’ve made real progress to solve this. We’re the furthest ahead in building something that’s sustainable and scalable; that will work in the existing landscape now, and as it’s changing. But we still need a lot of help. And we need the extraordinary kind.

We’re hiring in these categories (details below):

• Research Engineers.
• Cloud Architects and Software Engineers.
• Research Scientists (science ~= creativity) in ML, NLP or speech/audio/signal processing/etc.
• MDs and science PhDs that are top 0.01% communicators (high empathy) and read scientific papers on Sunday mornings.
• Advocates, health care workers and senior academics who didn’t sleep for two nights after reading this post.
• Research Interns, just as awesome as the Research Engineers/Scientists but still doing their master’s/PhD.
• Neuroscience students or recent grads with exceptional scientific writing skills ready to take ownership over entire clinical/academic projects.

You (it’s great to meet!)

For most of these roles we need people who are exceptionally technical and unconventional. Who solves impossible problems. Who table-flip the checkers board when losing “because it’s a stupid game requiring no strategy” and proposes to switch to Go (chess is sooo February 1996). Ideally someone who tries to make up new rules.

You need to have an appetite to work on fundamentally hard problems; a the-glass-is-overflowing type of person who laughs at the 9th failed attempt.

We want people who grow naturally (professionally and more importantly as human beings), that we can grow, and who can grow us. A+++ but we actually mean it.

Most important: you need to have a good reason for being here and wanting to work with us.

Research Engineers and Research Scientists

Why do we need you now?

We’re already producing by far the best-in-class research in NLP, speech/language understanding etc for speech changes in disease and the major bottleneck for us is capacity — having more hands to do great ML implementations here. That said, we also know there are many even better ideas out there we haven’t thought of, and need people even smarter than us to help us work out what those are.

What do we want you to be?

• Bursting with ML ideas (particularly deep signal processing, NLP, audio and text representations — and things we don’t know yet are important; tell us). Every time you read a new paper (which is often) it sparks a rich range of applications, related ideas, or combination of ideas.
• Natural communicator of complex concepts. Can read an ML paper and explain it to your grandparents in an engaging way (including grandparents that didn’t work on neural net models in the 70s).
• Obsessive doer that executes well. Someone who feels they always need to code things up that they read in papers (or better — code something up that’s better or shows how the paper is limited). Hungry for data.

What will you actually do?

• Read mountains of cutting-edge NLP/DL/signal processing papers, synthesize research and generate ideas (= application of ideas to our domain), synthesis of multiple ideas or entirely new ones. This could be papers such as:

Towards Robust Image Classification Using Sequential Attention Models

Continuous Hierarchical Representations with Poincaré Variational Auto-Encoders

Zero-shot Word Sense Disambiguation using Sense Definition Embeddings

Ordered Neurons: Integrating Tree Structures into Recurrent Neural Networks

Audio-Linguistic Embeddings for Spoken Sentences

• Have ongoing relationships and discussions with the best academics in the field to push the mutual research agenda.
• Code up and try out lots of ideas quickly.
• Take ownership of your research. Present your work to the rest of the team. Bring the best ideas to proof-of-concept or production level. If your idea is promising, lead a project and organise the other REs.
• Be a conscientious engineer when it counts (i.e. all the time except for quick mock-ups of ideas): minimizing technical debt through proper version control, documentation, reusability etc. And you should automate as much of this as you can.
• Scientific communication: writing papers, material for IP filings, making plots/graphics for customers.
• Don’t work in isolation, use the rest of the R&D team for ideas sharing, problem solving etc. Do your part to make this an academic community, but one that gets things done like a tech startup.
• Be extremely structured about the research/project management. This is the only way we can make research go fast without drowning. Conducting literature reviews, change logging, benchmarking etc.
• AUTOMATION AUTOMATION AUTOMATION. You will contribute to making everything we do as automatic and as interpretable as possible. CI/CD pipelines, automatic benchmarking of models etc. This will be a core part of the job. We are obsessed with automation and visualisation.

Must-have requirements?

• Degree from top university in comp sci/ML/maths/physics/etc. Or have done something that demonstrates equivalent (or greater) talents and skills.
• 3+ years of research experience in ML or applying ML, preferably for audio/linguistics/text (academic or industry, PhD great).
• Very strong coding skills in Python, which will be probed in the interview process.

Cloud/DevOps/Automation Engineers

Why do we need you now?

• Compliance, privacy, and data security are big deals to us. We need more expertise in our team building and scaling secure, compliant infrastructure for handling patient data. We’re doing this in the right way. Our infrastructure is a key selling point for us, rather than something that’s simply evaluated as a security risk. And that’s hard and takes a lot of work. We need this to be the best work you’ve ever done.
• We’re growing and scaling up fast. This includes commercial partnerships and running large clinical studies, where we’ll have much more data to be good protectors of.
• We’re specifically not contracting this work because this is something that will be constantly evolving, needing constant maintenance, and is mission-critical.

What do we want you to be?

• Experienced. A lot. This is vocational expertise. You can’t learn this from a book/degree. The ‘older and wiser’ the better. This is 1000% important.
• Opinionated. You need to have firm opinions everywhere you see it important for security/uptime/etc. Especially if anyone on the tech team asks to do something you’re against. We’re a startup, moving and scaling fast, so this will happen. You need to be able to shout at the CTO if need be.
• We’ll make unreasonable requests, both in terms of resources, time, feasibility — it’s up to you to stand your ground.

What will you actually do?

• (Jointly) own the infrastructure — design, build/scale, maintain it; with our team and the help of consultants/contractors if need be.
• Decide what resources we need (money, internal/external staff) and make a case for/request them.
• Fight fires: be on call if things go wrong with e.g. the backend during a clinical trial. Once we’re deploying live products at scale, on-call duties will be split between the CTO and any other cloud architect/full-stack people on the team. We share the good stuff and share the boring stuff.
• Brief, discuss and educate the CTO so they can make the best decisions and talk confidently to customers, investors, etc.
• Make decisions on our tech stack jointly with the CTO and other members of your team.
• Be extremely involved in the hiring process for non-research tech hires: design and lead algorithms interviews, intro calls, provide your opinion on candidates to the talent team.

Must have requirements?

• 5+ years in industry as cloud/DevOps/automation engineer. Big company, small company — we don’t care. You just have to be great.
• Spent your career mostly dealing with security/compliance (healthcare, fintech, etc.).
• Did all of this on AWS.
• This is recent experience. AWS changes a lot and keeping up is hard. It doesn’t count if you’ve been out of the full-stack game for years.
• Very experienced with Java, Python.
• Anything to do with product/UX/React Native/web dev/app dev is nice, but not a requirement.

MDs and science PhDs — Special Ops.

Why do we need you now?

• We’re here to win. We know how to win. You’ll be an important part of this. That’s why we need you.
• Data is important to winning our game. The research landscape is decentralised today with many small institutions running their own studies. We believe there’s power in bringing these smaller institutions together and working with them. More details for late stage candidates.
• Moving fast on this is core part of the CEOs high level strategy, so we need you right now. This is a Special Ops. role and you’ll be working directly with the CEO to execute on this.

What do we want you to be?

• This one is hard. Honestly we’re not sure what we need. I’m putting this out because of the asymmetric returns if we happen to get this right for the candidate. But you might know better, who we’re really looking for.
• Credible to doctors and leading professors. Highly professional.
• In depth, serious scientific discussions with same. Reads scientific papers on Sunday mornings.
• You probably need to be 1:10,000 top communicator/interpersonal skills (high empathy).
• Exceptional in execution.
• “Win-win” negotiator (you’ll be assessed on this at multiple abstraction levels through the interview process).
• Into how software is eating biology and healthcare. You need to have some conceptual understanding of how ML works, but don’t need to be a hands-on coder.
• High G-factor. This role has high uncertainty/fluidity/complexity and we know it will change, sometimes weekly.
• This is our role with the highest requirements; but if you’re the right candidate, you will have one of the most challenging and exciting roles you could dream of.

What will you actually do?

• Something not listed here within the first week.
• But: finding and obtaining access to existing datasets.
• Building relationships with academic groups, hospitals (or, more precisely, building relationships with the right stakeholders there).
• Brokering IP and licence agreements.
• Travelling. Sometimes not, sometimes a lot. Sometimes within the UK, sometimes out of.

Must have requirements?

• MD/PhD in neurology or something else very relevant.
• 1:10.000 top communicator.
• Experience with IP/licencing (candidates with this strongly preferred — but if you’re experience is the equivalent of the first Fosbury Flop for jumping over the bar we’ve set here otherwise, it might still be worth getting in touch).
• You might have moved from Mayo Clinic to Google Health; then a fast riser in companies such as Flatiron Health, Owkin, Tempus, Prognos, Linguamatics, HortonWorks, IQVIA.

You’ll work with

We’re a small, but fast growing, driven team of Oxford/Cambridge/Stanford machine learning researchers, neuroscientists, clinicians, and the exCEO of a top pharma company, backed by some of the best venture funds in Europe. We’re also a team of humans. Humans who are ambitious, humble, atypical, brilliant, and here to get to work to solve some fundamentally hard problems. If we succeed the rewards will be almost inconceivable. And we come to work every morning (or afternoon depending on circadian rhythm) with a deep sense of meaning. We’re many more things, which you’ll experience if you come and talk with us.

Speak with us!

All positions are listed on our job board here. It takes 1 minute to apply. Please get in touch.

Also share this with everyone you know who’s brilliant, empathetic, atypical. The people you know who came to mind as you read this.

“… we tend to forget that Icarus was also warned not to fly too low, because seawater would ruin the lift in his wings. Flying too low is even more dangerous than flying too high, because it feels deceptively safe.” — Seth Godin, oddball in the greatest sense of the word.

Image credits: Image 1 — Swissinfo.