In each of the last 25 years IBM has been granted more patents than any other IT supplier, putting it at the forefront of those actively designing future computer technologies. It has a string of 12 research labs around the world staffed by around 3,000 engineers and scientists well recognised for their activities (including 6 Nobel laureates). I travelled to Zurich to spend a day looking at a few of the research projects in a meeting chaired by Haig Peter. Some of their work-in-progress is difficult to grasp; quantum computing for instance is a mind-blowing subject – it seems the more I read, the less I understand! Nevertheless as an analysts it won’t stop that from attempting an overview.
Neural Networks, AI, machine learning – making chemical reactions easier to predict
Understanding and predicting organic chemical reactions has been a hand-made process in the past. IBM’s research student Theophile Gaudin showed us a new fully data-driven technique he’s been working on for his doctorate. By converting conventional chemical symbols and diagrams into text he’s been able to apply AI running on a neural network similar to modern systems used for computerised natural language translation. By ingesting hundreds of thousands of (textified) chemical equations the system has begun to grasp the ‘language’ of organic chemistry, allowing it to begin to predict the outcome of chemical reactions. In his published research Theophile has demonstrated results which were 80% accurate – similar to earlier non-computerised predictive methods. Although there’s some way to go to perfect the new system (adding the temperature changes during reactions for instance), it’s clear that the system can help in a number of ways – enabling young chemists to verify other findings for instance.
I think musical notation is another area in which conventional symbols could be converted into text for advanced machine learning; it might put paid to the accepted principal that artistic appreciation is mainly a subjective activity.
Secure permission blockchains for business
Spending on blockchain applications is growing massively in 2018 as IBM and other enterprise IT suppliers stretch their use beyond crypto-currencies into business use; Dr Andreas Kind discussed this subject in a presentation. Hyperledger – the Linux-based open source project which IBM is a part of – develops ‘permission blockchains’, which are unlike BitCoin in allowing data about transactions to be kept private between participants (consortia of partners, suppliers and customers) rather than broadcast to the world. IBM’s blockchain designs allow partial access to part of the blockchain for associate partners such as auditors, tax authorities and regulators.
IBM’s examples of recent blockchain customers include:
- Mearsk whose global shipping business is responsible for 90% of the world’s container freight movements. Currently a single shipment involving as many as 30 separate entities is governed by a paper-based process in which the loss of a single document can hold things up for months.
- Another is UBS’s automotive insurance business which is developing a ‘transport as a system’ model to address the growth of shared vehicles, where the car itself, rather than the driver, has to pay for fuel, parts and servicing.
In both cases a secure form of authentication is needed for physical objects (containers for Mearsk, car parts for UBS) are genuine. The challenge is the prevalence of counterfeiting (as much as 50% of aftermarket car parts are fake in some countries for instance), which is likely to become easier as the processes they’re involved with become more automated.
Crypto anchors – chemical encryption for secure built-in provenance
In his presentation Dr Emmanuel Delamarche told us that counterfeiting is also rife in the Healthcare industry; for instance in India two people were charged with distributing counterfeit HIV/AIDS testing kits. Clever forgers have even been able to copy the holographic labels manufacturers add to packaging to prove provenance.
Blockchains are often more secure than the products and offerings transacted through them and no one wants a system that proves the provenance of a counterfeit item. To address this IBM research has been working on Crypto Anchors – adding unique identifiers to products proving them to be genuine through use.
IBM first added ‘ultra-complex transient optical security patterns’ into microfluidic chips for diagnostic test machines before retrofitting them to rapid diagnostic test kits used by patients at home. A series of coloured dots is printed using an inkjet printer onto the test strip allowing the providence to be proved using a smart phone app. Embedding the crypto anchor into the product itself, which is validated by the test itself, makes them far harder – if not impossible – to copy.
There are many other potential ways of designing crypto anchors in other areas of course.
Neuromorphic Computing – get memory to do the processing
In his presentation Dr Abu Sebastion explained that the human brain is very different from most computers in use today; for instance the mesh of Power processors in the Watson system used to win Jeopardy! used about 800kW of energy to run in competition with humans whose brains take the equivalent of just 20W. Although no one yet knows how the human brain works in detail, we do know that memory and processing are combined, allowing memory to take centre stage. In its research labs IBM is working to modify conventional von Neumann computers to make them more like the brain – hence the term ‘neuromorphic’. In particular it’s using new storage class memory components to build ‘computational memory’ to emulate the synapses of the brain. Unlike computer memory in use today where data is held by changes in on-chip capacitors, new ‘resistive memory devices’ hold data on metal-oxide phase-changeable material. If successful the new machines will be better for deep machine learning, cognitive and AI applications than today’s neural network machines which can take many weeks to ‘train’.
Quantum computing – IBM Q and experimental qubits
The most interesting – and most difficult to understand – topic introduced by Dr Stefan Filipp was quantum computing, which will has the potential to create orders of magnitude improvements in computing over current systems if/and when they can be made to work.
All current computers are digital machines, converting all data into bits, which have only 2 binary states – 1 or 0. Quantum computers are based on qubits, which have many more (see the Bloch sphere in the Figure). Quantum computers are difficult enough to understand theoretically, making an operational one is even harder; and yet IBM (one of only a handful of companies and organisations in the world with appropriate resources) is working at it. We were shown IBM Q in the labs – a working quantum processor with super-cooled elements (see Figure) needed to shield the processor from external interference. Nevertheless the current system is only operational for milliseconds.
The creation of workable quantum computing will reduce the number of mathematical problems which are currently too complex to be solved. It’s highly likely that they will be used by criminals and security agencies to crack many of the encryption codes in use today. As one of a handful of research organisations building prototypes IBM should be able to make money from building and selling quantum computers, and entire data centers of conventional computers needed to input and extract data from them and new forms of encryption which will be harder to crack than those in use today.
Post Quantum-Crypto – an urgent need to replace current encryption schema
In his presentation Dr Michael Osbourne talked to us about the consequences quantum computing will have on encryption systems. He suggested that there is no such thing as an unbreakable key, just ones that haven’t yet been hacked; for instance there are already companies claiming to be able to break 512-bit keys – a process which takes 4 hours and costs about $75.
Vital protected data legally has to be securely retained for many years – HIPPA (6), taxation (7-10), ‘guide 0068’ chemical trials (25), etc. New laws, such as the French government’s to make manufacturers responsible for the security of their products throughout their lifecycle until the last one is retired, are expanding the need for long-term retention of critical information, especially as we embed more computing and data into devices such as appliances, cars. It’s clear that confidentiality is important for governments; for instance the US is still keeping secret its findings into who shot president J F Kennedy.
Data will become increasingly difficult to protect with the advance of quantum computing. To help the US National Institute of Standards and Technology (NIST) has begun to address Post-Quantum Cryptography (PQC) standards with the first draft planned for 2023. The need will become urgent if and when the number and predictability of quantum computers increases. I expect IBM to launch its own PQC offerings sometime in the future.
IBM research is working on Lattice-based cryptography, which hides data within complex mathematical algorithms (lattices), which should prove more resilient to attacks from quantum computing. These should extend secure data retention timescales massively. Lattice-based cryptography is also the basis of IBM’s Fully Homomorphic Encryption (FHE), which will make it possible to perform calculations on a file without being able to read sensitive data or exposing it to hackers.
Creating the future – a hard job, but someone has to do it!
Most IT suppliers’ success is based on eking out the value of existing components, protocols and software and pushing the transition from one technology to another. For them it’s often harder to predict ‘when’ than ‘which’ new systems will succeed; nevertheless the future of IT won’t create itself. IBM is one of only a handful of suppliers using a deep, long-term commitment to scientific research to investigate and create genuinely different types of technology. It’s working to deliver vast (sometime scary) improvements in the speed and capacity of future computer systems as well as to defend us from their malicious use. It will make money from its inventions and the mass of conventional computing needed to get some of it to work. As patent holder it will also make money once these technologies go mainstream of course.