Hello, and welcome to the HORIZON weekly newsletter. Particularly warm greetings to our many new subscribers - please do forward this on to colleagues and connections in your network who would also enjoy the insights.
Below you will find some hand-picked fresh thought-leadership content, giving you an overview of recent developments, topical innovations, and what we're seeing and hearing out there towards the digital frontier.
We have some slots available for Q1 engagement - drop us a line now to discuss by replying directly to this Email.
Thank you for reading and being a part of the HORIZON community.
If this Email has been forwarded to you, you can access previous editions and sign-up to receive future instances for free at: https://future-horizon.kit.com/posts
Recent articles
|
|
Making You Smarter: "Neural Processing Unit (NPU)" explained.
Let's start with a definition.
A NPU is a specialised type of computer chip with an architecture that mimics a human brain's neural network.
In a nutshell: with design focused on accelerating computations, NPUs drastically enhance the speed and efficiency of Artificial Intelligence (AI) systems.
NPUs are purpose-built for problem-solving functions and can improve over time, learning repetitively from different types of data and inputs.
They are specially designed to process Machine Learning (ML) algorithms - the foundation upon which AI applications are built.
Processes large amounts of data in parallel, a NPU can perform trillions of operations per second.
NPUs are particularly efficient at tasks like image recognition and natural language processing, which require fast processing of massive amounts of multimedia data.
Importantly, as well as being far more efficient at specific AI tasks than a CPU (Central Processing Unit) or GPU (Graphics Processing Unit), it uses less power.
As the number of AI operations on our devices increase, having a NPU can free up the CPU and GPU for other tasks that they are best at - all three are complimentary.
In time, NPUs at scale could revolutionise the face of AI/ML workloads - unlocking significant business value.
NPUs work well for real-time inference tasks in on-device AI processing, plus also in edge hardware where low latency and energy efficiency are key.
In the near future, as you interact with your device(s) more with conversational voice it will be the NPU in the background which infers your input for an AI agent to act upon.
Another use case for incorporating NPUs into consumer-grade electronics might be an AI-enabled camera which works better at blurring the background on a video call (thanks to improved image recognition).
NPUs are likely to become increasingly popular in TVs (which traditionally haven't had advanced processors), where they can upscale the resolution of older content to more modern 4K definition.
Many leading hardware manufacturers are actively working on NPUs, including NVIDIA, Intel Corporation, Microsoft, Samsung Semiconductor, AMD, Apple, and Qualcomm.
Even with recent market jitters thanks to Chinese firm DeepSeek AI, NPUs still have a place as they focus on inference; DeepSeek has primarily caused a frenzy due to its alleged reduced cost of GPU training.
In late 2023, the term "AI PC" was coined for computers with next-generation processors; fundamentally, an AI PC is a machine that performs AI processing locally (specifically via a NPU).
Some PCs and laptops are already designated "Copilot+", meaning they can run Microsoft's Copilot AI on an onboard NPU.
Thus before long, as this emerging tech become more ubiquitous your new laptop or mobile device may well sport a NPU - so keep an eye out.
|
|
|
2025: the year to become Quantum-ready.
Not just according to Future Horizon, but also tech hyperscaler Microsoft.
"We’re seeing a growing need for business leaders to have the information and tools to understand better the depth of these amazing technical breakthroughs and the business applications and value they open up".
The above quote comes from a recent blog post by the President and Chief Operating Officer of Strategic Missions and Technologies.
2025 has been anointed the International Year of Quantum Science and Technology (IYQ) by the United Nations.
Two specific recent advancements have made the Quantum emerging future horizon a little closer.
Researchers at Massachusetts Institute of Technology have achieved a breakthrough which marks a notable step towards the realisation of practical Quantum computing.
Qubits - the building blocks of Quantum computers - are highly susceptible to noise and control imperfections.
The team used timed pulses and a form of synthetic light to reduce errors and improve accuracy - setting a new world record with 99.998% fidelity.
Elsewhere in the USA, another group has achieved the first uninterrupted Quantum entangled signal over a commercial fibre optic network in a significant security milestone.
Researchers from The University of Tennessee at Chattanooga and the Oak Ridge National Laboratory continuously sent signals for 30 hours between nodes setup ~800m away from each other.
Future Quantum internet networks are likely to be more secure and capable than existing networks we have today.
December 2024 saw major announcements from IBM (introducing a 1,121 superconducting qubit Quantum processor called "Condor") and Google (revealing a new chip called "Willow" which reduces errors exponentially as more more qubits are used).
Willow raised eyebrows as it performed a standard benchmark computation in under 5 minutes that would take one of today’s fastest supercomputers 10 septillion years - a number that vastly exceeds the age of the Universe.
Not all Big Tech is necessarily sanguine about Quantum in the near term: NVIDIA CEO Jensen Huang’s stated at CES 2025 that “very useful” Quantum computing is still 20 years away.
In China, two Tech firms - Baidu, Inc. and Alibaba Group - withdrew from pursuing Quantum; this is due to a macro major difference in approach.
Whilst efforts in the West are a competitive mix of startups and Hyperscalers, the Chinese model towards Quantum advancement is very much directed at the State level, with the government controlling funding top-down.
The current short-term potential of Artificial Intelligence (AI) is ambitiously overhyped; its ROI demonstrably elusive - a true Quantum breakthrough will make Gen AI look positively pedestrian.
|
|
|
A concise message for business leaders: Artificial Intelligence (AI) products will never be secure.
Researchers at Microsoft have said it is an impossible task.
26 "Red Team" experts tested more than 100 Microsoft AI products, concluding that the key to AI security is raising the cost of attacking, using three specific methods.
Quote: "...the idea that it is possible to guarantee or “solve” AI safety through technical advances alone is unrealistic and overlooks the roles that can be played by economics, break-fix cycles, and regulation."
Also: "The work of building safe and secure AI systems will never be complete."
No system - AI or otherwise - can ever be guaranteed 100% secure.
The cybersecurity challenge is never static: it has been and always will be a constantly morphing contest between cops and robbers regardless of what the underlying tech is.
So what makes AI different when looking through a security lens?
While it can be a useful tool for analysing code and finding vulnerabilities, AI has been both a blessing and a curse for cybersecurity.
When it comes to Large Language Models (LLMs) - the engine of much which is being sold as AI - these amplify existing security risks and introduce fresh ones.
Non-deterministic algorithms are "black boxes" - nobody knows precisely how they work (or what they will generate as output); inconsistency makes providing adequate and appropriate defence challenging.
Adversaries are also using AI creations as active weapons: sophisticated attacks are accessible and cheap.
Good news for cybersecurity professionals - and lawyers - but sobering for everyone else both today and as the tech evolves further in the future.
Microsoft - and frankly every other tech company both established and startup - is crowbarring AI into every bit of software it can - so it can raise its prices.
Organisations must keep this top-of-mind when considering what data and operations to hand-over for automation by much-fabled "agents".
Much is being made of Agentic AI and how much more cost-effective and efficient it will make firms; Salesforce is already sacking expensive warm fleshy bipeds who they are replacing with bots.
Realistic expectations by peddlers of products seemingly remain in short supply, especially when it comes to real-world risks.
AI is not the apocalypse, but it isn't the pot of gold at the end of the rainbow either - no tech ever can be.
Reply to this message if you want a link to the paper detailing the findings of the Red Team.
|
Thank-you for reading and being part of our community - we trust you find these original pieces on emerging technology and digital innovation useful, valuable, and thought-provoking as we bridge the gap between today and what future tech might bring tomorrow in Plain English.
When you're ready, contact us to discuss how we can deliver independent, objective, and unbiased strategic foresight around the implications of emerging technologies for your organisation -
https://www.futurehorizon.digital/
Think bold.
Think broad.
Think beyond.
If you would like to explore sponsoring an edition of HORIZON, please get in touch at horizon-sponsorship@futurehorizon.digital
If you are not currently a retained client but value the weekly insights in HORIZON and wish to support its ongoing work, you may kindly consider contributing using the Buy Me A Coffee page: https://buymeacoffee.com/futurehorizonco
As ever, we welcome all forms of feedback: compliments as well as constructive criticism! If there are particular topics you want to see more - or less - of, please let us know. You can reach us at horizon-weekly@futurehorizon.digital
