Wiser Technology Advice Blog

HOME
WISER-TECHNOLOGY-ADVICE-BLOG
NAVIGATING THE FRONTIER OF GENERATIVE AI

Navigating the frontier of generative AI

10 December 2023

Risks and rewards of generative AI, with some recent developments

Generative AI is a rapidly growing field that has the potential to revolutionize the way we live and work. However, as with any new technology, there are both risks and rewards associated with its use.

In this blog post I will discuss the risks and rewards of generative AI, providing an overview of its implications and a few potential uses. I’ve also included some interesting recent developments about Microsoft and Google’s offerings in this space.

Generative AI in a nutshell

ChatGPT brought AI to the public’s attention, but generative AI is not just limited to written language interactions using large language models (LLMs). It is also capable of generation of speech, images, video, music and computer code.

AI tools learn patterns and structures from input data and then create new data with similar characteristics. Generative AI tools can produce text, images, or other media based on what they’ve learned. Imagine them as digital artists, weaving new content inspired by existing examples.

successful generative AI examples — source: TechTarget

Written Language Interactions:

Chatbots and Virtual Assistants: Generative AI can power conversational agents like chatbots, with output presented in an accessible tone. These tools use large language models (LLMs) to interact with users and generate content. Chatbots like ChatGPT, Copilot, and Bard accept natural language prompts and respond with coherent text, engaging in human-like text-based conversations.

Content Generation: Generative AI LLMs can create articles, stories, or summaries based on prompts, for example personalised news updates, short stories and poetry.

LLMs can be used in business for summarising documents, writing customer-facing materials, and explaining complex topics in natural language.

Speech:

Audio interaction: Much like written language outputs, audio can be generated by AI in natural, conversational, and even colloquial styles with the capacity to rapidly shift among languages, tone, and degrees of complexity.

Text-to-Speech: Generative AI tools can convert written text into natural-sounding speech. Examples of this are voice assistants, audiobooks, and accessibility tools.

Voice Synthesis: By training on voice recordings, generative AI can mimic specific voices, making it useful for dubbing, voice acting, and personalized voice messages.

Images:

Image Generation: Generative AI tools like Stable Diffusion, Midjourney, and DALL-E create stunning visual art from textual descriptions. For example, you can describe a “fire-breathing dragon in a snowy forest,” and DALL·E will create an image matching that description.

Style Transfer: Generative AI tools can transform images by applying the artistic style of another image. This technique is used in photo filters and creative visual effects.

3D representation: From text or two-dimensional images, AI tools can extrapolate and generate data representing 3D objects.

Examples for business use of image generation include simulating how a product might look in a customer’s home, reconstructing an accident scene to assess insurance claims and liability, and AI-assisted prototyping and design in a purely virtual space.

Video:

Video Synthesis: Generative AI tools can create realistic video sequences. These tools can take user prompts and output videos, with scenes, people, and objects that are entirely fictitious and created by the AI tool. Applications include video game graphics, special effects and ever more convincing deepfake videos.

Video Captioning: Generative AI tools can generate descriptive captions for video content, enhancing accessibility and searchability.

Examples for business use include autonomously generating marketing videos to showcase a new product and simulating dangerous scenarios for safety training.

Music:

Music Composition: Generative AI tools can compose original music. They learn patterns from existing compositions and generate new melodies, harmonies, or entire pieces.

Remixing and variation: Artists can use generative AI tools to remix existing tracks, creating fresh versions or mashups.

Computer code:

Code generation: Generative AI can write code snippets based on high-level descriptions. Computer code can be developed using generative AI tools in a variety of programming languages, with the capacity to autonomously summarize, document, and annotate the code for human developers.

Bug detection and repair: Some generative AI tools can identify and fix code errors, improving software quality.

Risks of generative AI

Generative AI is still an emerging technology, so you must be careful when implementing it to avoid any negative consequences.

deepfake image prompts — source: Metaphysic

Some of the biggest risks of generative AI concern trust and security, including hallucinations, deepfakes, data privacy, copyright issues, and cybersecurity problems.

If you’re going to introduce use of generative AI in your organisation, be sure to educate everyone in the team about its risks and put guardrails in place for its use. Ensure there’s a human in the loop, test and re-test, and get feedback from people on how well the tool has been performing.

Accuracy and reliability of results

While AI is getting smarter all the time, it will never be perfect.

Generative AI providers cannot guarantee the accuracy of what their algorithms produce, nor can they guarantee safeguards against biased or inappropriate content. Inaccuracies can have serious consequences for decision-making, spread of false information, privacy violations, legal liabilities, and more.

Further, AI tools can have hallucinations, where the AI perceives patterns or objects that are non-existent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate. An analogy for AI hallucinations is how humans sometimes see figures in the clouds or faces on the moon.

Cyber security

One of the biggest risks of generative AI is cybersecurity. With the power to generate content, there is an inherent risk of misuse. From deepfake videos to manipulated text, generative AI can be exploited for malicious purposes, raising concerns about misinformation and cyber threats.

Through generative AI, attackers may generate new and complex types of malware, phishing schemes and other cyber dangers that can avoid conventional protection measures. Such assaults may have significant repercussions like data breaches, financial losses and reputational risks.

Data security and privacy

Publicly available generative AI tools such as ChatGPT use content from across the internet and gather up and re-use any information you provide when posing questions. Confidential and/or sensitive information should never, ever be shared with a publicly available tool. A breach could not only compromise your organisation’s confidential data but also erode trust with stakeholders.

Ethical and legal concerns

Generative AI systems can inadvertently produce biased or discriminatory content, reflecting the biases present in the data on which they were trained. These biases may perpetuate existing inequalities, posing a challenge to ensure fairness in AI applications.

Generative AI tools indiscriminately hoover up training data from across the internet, with no respect for privacy or copyright of the information. This means you may not be aware that content it creates has infringed on copyright laws, which can lead to legal liabilities and financial losses.

The increasing popularity of AI and the huge number people using it is leading to a growing environmental footprint. Gartner predicts that by 2030 AI could consume up to 3.5 percent of the world’s electricity.

Recent developments

There have been some interesting developments in recent days in the race for dominance of generative AI tools.

Google Bard demo video

Alphabet’s generative AI offering is Google Bard, a chatbot that can answer questions and generate text from prompts. It was trained on a specific dataset for conversations. Google Bard is free, with an unlimited number of questions.

Google Gemini blue duck — source: Google

The initial release of Google Bard was widely panned by critics, but with its latest release it seems to have leapfrogged the capability of other tools and finally be living up to expectations. Google has unveiled Gemini, a new multi-modal engine behind Google Bard that can recognise and reason from a combination of audio, visual, and text-based input, giving Bard a real boost in capability.

Google promoted the release of Gemini with impressive demonstration video, showing how the multimodal mode of speech and image interactions works. Watch the video here: Hands-on with Gemini: Interacting with multimodal AI - YouTube

To begin with, it narrates an evolving sketch of a duck from a squiggle to a completed drawing, which it says is an unrealistic colour, then evinces surprise ("What the quack!") when seeing a blue rubber duck. It then responds to various voice queries about that toy, then the demo moves on to other show-off moves, like tracking a ball in a cup-switching game, recognizing shadow puppet gestures, reordering sketches of planets, and so on.

Just one problem: the video isn't real.

Although the Gemini AI engine might be able to do the things shown in the video, it didn't, and maybe couldn't, do them live and in the way it implies. The video was actually a mash-up of carefully tuned text prompts with still images. The demo video was edited and shortened to misrepresent what the interaction is actually like.

How embarrassing. The new Gemini AI engine behind the Google Bard tool might be great, but Google have done themselves a disservice by putting out a misleading video.

Microsoft’s Copilot branding is confusing

I find the Microsoft brand ‘Copilot’ really confusing.

I’d originally understood Copilot to be part of Microsoft 365 and expensive to access, only available at an additional fee for organisations with Microsoft Enterprise level 5 licences.

source: Microsoft

But Microsoft has recently launched Copilot in the Bing internet browser, which is freely available to anyone with a Microsoft account.

And there is a third AI tool using the Copilot brand, Github Copilot. This is integrated with Microsoft Visual Studio and is used to assist with writing code and generating technical documentation.

After some digging around, I have learned that Microsoft 365 Copilot is meant to be an AI assistant for productivity and the new Copilot in Bing is intended top help you find answers online.

Microsoft 365 Copilot:

Purpose: Designed to enhance efficiency of Microsoft 365 applications and services, Edge, Microsoft Bing, and Windows.
Capabilities: Leverages the advanced capabilities of OpenAI’s GPT-4 large language model (LLM). It can generate responses based on the context of your organisational data, such as user documents, emails, calendar, chats, meetings, and contacts, to improve productivity and efficiency.
Integration: Available in Teams, Word, Outlook, PowerPoint, Excel, and other Microsoft 365 apps.
Training data: Information from within your organisation’s Microsoft 365 tenancy.
Security: Enterprise-grade security, privacy, compliant with existing privacy, security, and compliance commitments to Microsoft 365 commercial customers, including the General Data Protection Regulation (GDPR) and European Union (EU) Data Boundary. Your business data is used to improve context only for your use, and the LLM itself doesn’t learn from your usage.
Pricing: US$30.00 per user per month, only available with a Microsoft 365 E3 or E5 licence.

Copilot in Bing:
- Purpose: A chat interface (formerly Bing Chat) to provide content generated from information found across the internet.
- Capabilities: Copilot leverages the advanced capabilities of OpenAI’s GPT-4 large language models (LLMs). It does NOT have access to your organisation’s data or content within Microsoft 365, it is trained on publicly available data gathered from across the internet and therefore tends to have more generalized results.
- Training data: Publicly available information from across the internet.
- Pricing: free with a Microsoft account, such as an outlook.com or hotmail.com email address and password or the login information you use for Microsoft services, such as Office, OneDrive, or Xbox. Microsoft Copilot in Bing is the only way to use GPT-4 for free at this time, and Microsoft claims the integration with the latest language model makes Copilot more powerful and accurate than ChatGPT. The free version of ChatGPT, GPT-3.5, is only trained on data up to the year 2021, so it cannot provide answers related to more recent events and developments.

I am concerned about the potential confusion between these two Microsoft Copilot brands.

Users of Microsoft 365 Copilot are guaranteed that their organisational data will be kept safe, with all of the Microsoft enterprise level security measures in place. However, users of Copilot in Bing are given no such guarantees. By using this tool you are granting Microsoft and its partners permission to use the prompts and information you put into the chat interface, handing over full licence rights to your content.

How will users know the difference between these two types of Copilot when they are using Microsoft applications?

Benefits and use cases

AI has the potential to revolutionize the way businesses operate. One of the primary differences between more traditional AI and Generative AI is that the latter creates output that feels more intuitive to interact with.

AI chatbot — source: ZDNet, Carol Yepes, Getty images

Generative AI can help automate tasks and processes, freeing up time and resources for more complex and creative work. It can be used to help create new products and services, accelerating research and development through generative design.

For example, law firm Minter Ellison has recently set up its own version of Microsoft Copilot to assist in identifying productivity opportunities and drive innovation in the legal sector.

Marketers have known for a long time about the power of data, tracking customer interactions and personalising marketing messages to customers. With generative AI tools, it is possible to enhance customer experience through personalised interactions. But a word of caution: targeted marketing is more acceptable to younger generations, so be sure to understand who you’re marketing to and how much they care about their privacy.

When it comes to costs and benefits, there are several factors to consider. While some publicly available generative AI tools are free to use, others require a subscription or licensing fee. Some AI tools are more accurate and reliable than others, while others are more flexible and customisable.

There is a big difference between using generative AI for basic, low-risk tasks and applying it at an enterprise level. The nature of publicly available generative AI means answers can be inconsistent and unpredictable, which may be acceptable in standard consumer use, such as checking a store’s opening hours. However, if a bank or financial provider is using generative AI to respond to queries about policies or new products, there is no room for error.

Generative AI has a wide range of applications across various industries. Here are a few use cases to get you thinking about how it could be used to benefit your business:

Chatbots and Virtual Assistants:

Generative AI powers intelligent chatbots and virtual assistants, enhancing customer support, automating responses, and providing personalized interactions.

Design and Development:

Generative AI tools can assist in designing products, creating architectural layouts, and generating visual assets. For instance, they can generate logo designs or website layouts.

Content Creation and Repurposing:

Generative AI can create articles, summaries, and marketing content. It can also repurpose existing content into different formats (e.g. turning a blog post into a video script).

Data Analytics:

Generative AI toolscan synthesize new data points, augmenting datasets for training other AI tools. They can also generate synthetic data for privacy-preserving analytics.

Risk Mitigation:

In finance and insurance, generative AI can simulate market scenarios, stress tests, and risk assessments. It helps organizations prepare for potential risks.

Predictive Maintenance:

By analysing historical data, generative AI tools predict equipment failures or maintenance needs. This is valuable for industries like manufacturing, energy, and transportation.

Quality Control:

Generative AI can identify defects in manufacturing processes by comparing real-world data with expected patterns. It ensures product quality and reduces waste.

Inventory Management:

Optimizing inventory levels is crucial for retail and supply chain management. AI tools can predict demand fluctuations and recommend inventory adjustments.

Customer Interactions:

Generative AI enhances personalised recommendations, email marketing, and targeted advertising. It tailors content based on user behaviour and preferences.

Novel Design Exploration:

AI tools explore multiple design possibilities, helping engineers and architects find optimal solutions. They can invent novel designs that humans might overlook.

Want to have a chat?

Despite the risks and concerns of this fast-moving space, I hope this blog post has inspired you to think about how to harness the potential of generative AI to drive innovation and efficiency in your organisation. In fact, I used Copilot in Bing to help write some (but not all) of this blog post!

With 35 years’ experience in the technology industry, I have the knowledge and skills needed to help you plan the transformation of your through technology. If you’d like to talk further about this or anything I’ve written about, get in contact with me today, I’m always happy to meet and have a chat over a coffee.

GitHub Copilot is generally available to all developers, 21/06/2022, Thomas Dohmke, GitHub, available at: https://github.blog/2022-06-21-github-copilot-is-generally-available-to-all-developers/#:~:text=We%E2%80%99re%20making%20GitHub%20Copilot%2C%20an%20AI%20pair%20programmer,students%20and%20maintainers%20of%20popular%20open%20source%20projects

Google debuts impressive Gemini AI model, 7/12/2023, David Braue, Information Age, available at: https://ia.acs.org.au/content/ia/article/2023/google-debuts-impressive-gemini-ai-model--.html

Google’s best Gemini demo was faked, 8/12/2023, Devin Coldewey, Yahoo! News, available at https://au.news.yahoo.com/googles-best-gemini-demo-faked-220636940.html

GPT-3.5 vs GPT-4: Is ChatGPT Plus worth its subscription fee?, 15/11/2023, Steven Vaughan-Nichols and Maria Diaz, ZDNet, available at: https://www.zdnet.com/article/gpt-3-5-vs-gpt-4-is-chatgpt-plus-worth-its-subscription-fee/

Hands-on with Gemini: Interacting with multimodal AI, Google, available at: https://www.youtube.com/watch?v=UIZAiXYceBI

How to use Copilot (formerly called Bing Chat), 17/11/2023, Maria Diaz, ZDNET, available at: https://www.zdnet.com/article/how-to-use-the-new-bing-and-how-its-different-from-chatgpt/

Microsoft 365 Copilot Cheat Sheet: Price, Benefits & Release Date, 6/10/2023, Mark W. Kaelin, TechRepublic, available at: https://www.techrepublic.com/article/microsoft-copilot-cheat-sheet/

MinterEllison sets up own AI Copilot, 4/12/2023, Kate Weber, ITnews, available at: https://www.itnews.com.au/news/minterellison-sets-up-own-ai-copilot-603200

Part 1 Generative AI special report: evaluating the risk vs rewards, 28/09/2023, Jon Stone, ProbeCX, available at: https://www.probecx.com/en-au/blog/generative-ai-special-report-part-1-evaluating-the-risk-vs-rewards

Successful generative AI examples and tools worth noting, 20/10/2023, John Burke, TechTarget, available at:https://www.techtarget.com/searchenterpriseai/tip/Successful-generative-AI-examples-worth-noting

The Future of Generative Adversarial Networks in Deepfakes, 25/07/2022, Martin Anderson, Metaphysic, available at: https://blog.metaphysic.ai/the-future-of-generative-adversarial-networks-in-deepfakes/

The Generative AI Dossier, 2023, Deloitte, available at: https://www2.deloitte.com/content/dam/Deloitte/us/Documents/consulting/us-ai-institute-gen-ai-use-cases.pdf

What is generative AI and why is it so popular?, 14/09/2023, Sabrina Ortiz, ZDNET, available at: https://www.zdnet.com/article/what-is-generative-ai-and-why-is-it-so-popular-heres-everything-you-need-to-know/

Why Is Generative AI Important for Business?, 24/08/2023, Bill Leigh, Salesforce, available at: https://www.salesforce.com/blog/generative-ai-for-business/