There is plenty of opinion kicking around so I thought join in on the conversation around artificial intelligence and cover some of the basics, the recent AI Safety Summit, and how I’m using the technology.

I believe artificial intelligence will surpass the internet's impact, becoming a major societal trend with transformative effects on our industry, and although the market is still in its infancy, its evolving quickly.

Dene Lewis, Chief Technology Officer, CAE Technology Services. 

The AI Safety Summit 

Bletchley Park was the perfect host for the first AI Safety Summit at the end of last year – I’m a history buff as well as a techie so recommend a visit. It was a high priority event, attended by senior global leaders from business, technology, and governments. The summit focused on two types of AI: 

• Frontier AI, general purpose AI that can perform a broad range of tasks 
• Narrow AI, models that focus on single or limited tasks 

 And two types of risk: 

• Misuse risks, for example where a bad actor using AI to create harm (e.g. cyber-attacks) 
• Loss of control risks that could emerge from advanced systems (terminator scenario, amongst others) 

When we consider risks of artificial intelligence, we tend to think Matrix or Terminator level existential threat to humankind, but more imminent risks such as AI in warfare, cybersecurity, and sustainable AI (AI growth = more data centres) also saw airtime. 

 My key takeaways 

• 28 countries signed the ‘Bletchley Declaration’ international statement on AI. 
• Various taskforces announced sets AI towards governance. 
• Major AI companies and governments agreed to collaborate on testing. 
• The Summit and media noise are driven by the rapid adoption of Generative AI tools like ChatGPT, Bard, and DALL-E. ChatGPT took two months to get to 100 million users, Instagram took 2.5 years. 

Generative “Gen” AI 

Two key terms you should know. 

• Generative AI - a general AI category that learns from huge amounts of data and creates new content like art or text. 

• Large Language Model (LLM) - a specific type of Generative AI that understands and generates human-like language, making it useful answering questions. 

I struggle with creating first drafts, so ChatGPT and Bing Chat helps me turn rough notes into drafts. I’ve also used it to create slide deck plans and apply business strategy frameworks to my thoughts. It has also helped me in personal plans.

Check out this link which shows how I planned my birthday present from my wife, a trip to Ypres battlefields (she of course, loved it!). 

Be careful using company or personal data in free GenAI tools. Use Bing Chat or similar as an enterprise-ready option with data privacy features . 

ChatGPT 

ChatGPT is an LLM and ‘AI’s killer app’ and will transform productivity. GenAI can build highly personalised, hyper-focused marketing campaigns that tailor your message to individuals at scale through AI automation. Now, change the word ‘marketing’ to ‘phishing’ and consider the misuse risks. Cybersecurity’s new frontline will be - AI. 

GenAI is not perfect yet. It can plagiarise or give wrong answers. My above example is limited and generic, however the potential for improvement is huge and the capabilities in 5-10 years will be mind-blowing. The internet went from dial-up and text-based sites to mobile and video streaming in under a decade. 

Where are we now? 

AI has evolved from Machine Learning and Deep Learning initiatives which developed over the last decade. I see similarities with early 2010s public cloud, when the technology moved from innovators and tech-savvy toward mainstream early adoption.  

Technology companies are already adding AI to their solution stacks. For example, Cisco is building capabilities by adding a generative AI policy assistant for Secure Firewall FMC. This will let IT admins use natural language to change or check policies. We expect more of this across the market, as AI will enable IT teams to troubleshoot and automate tasks more easily, improving productivity, reducing risk of downtime, and shrinking time to resolution. 

Cisco plans to add a generative AI policy assistant for Secure Firewall FMC 

Apps need Infrastructure 

AI tools are software applications with very specific infrastructure requirements such as: 

• GPU-dense servers to run AI software 
• High-performance, scalable storage for data to train models 
• Fast and secure networks 
• Software stacks for resource management 

AI hybrid cloud deployment options include but not limited to: 

• On-premises – build your own AI platforms (e.g. NVIDIA, Dell PowerEdge and PowerScale) 
• Public Cloud IaaS – run GPU servers and storage in the cloud (e.g. Azure AI Infrastructure) 
• Public Cloud PaaS – use ready-made AI tools (like Azure OpenAI Studio) 
• SaaS – use AI software-as-a-service (like ChatGPT, Copilot or other industry-specific offerings) 

AI will prompt data centre modernisation and the next wave of cloud migration, as AI-native apps are deployed or existing software products introduce AI capabilities. For example, the Healthcare industry use PACS systems to store medical images. Introducing AI capabilities could help automate elements of diagnosis and treatment plan creation, but they will need GPU-dense infrastructure to leverage these innovations.  

NVIDIA is the leader in GPU technology, which are ideal for artificial intelligence. NVIDIA was able to leverage its knowledge and experience of enterprise data centre technologies, (it’s been doing app acceleration for many years - think VDI) to build an integrated AI technology and software architecture. Their strategy transcends the GPU and positions them as the AI technology platform of choice. They already underpin every deployment option mentioned above, a reason why they are worth over a trillion dollars! 

Preparing for AI 

AI adoption will be use case-driven, and needed to demonstrate business value. Microsoft, who invested in OpenAI, is leading the charge for AI in user productivity with a flurry of Copilot offerings. While Microsoft 365 Copilot will see high demand from users, organisations must implement effective adoption plans to ensure the rollouts achieve ROI. 

Additionally, robust security and data governance policies are a must! Security by obscurity will melt away with GenAI apps, if you can access it then AI will find it! That said, vast amounts of data will improve model’s effectiveness, so organising data before adopting AI will increase time to value. 

Wading Through the Options 

The technology is there, but the ROI and TCO are key. At CAE, we can provide help you understand how to leverage AI technologies and navigate the vast array of options. We also offer workshops and data classification services underpinned our Methodology, to support your AI journey.