In the first half of 2023, we've witnessed an explosion of interest in generative artificial intelligence (AI). The reaction has been equal parts enthusiastic and cynical: for every claim that the technology will change society for the better by automating away menial busywork, we also hear that it will have dire consequences, ranging from worker displacement to the end of civilization.
So what should businesses and individuals make of these extraordinary claims? At Conveyance Marketing Group, we keep a close pulse on the tech landscape, and we are keenly aware of Gartner's Hype Cycle: when people start talking about a new invention changing the world - for better or worse - that is usually a sign of inflated expectations.
Going all the way back to the 1960s, AI in particular has provoked waves of excitement followed by periods of intense disappointment and stagnation (AI Winter). But every winter brings the spring - and spring is here.
The New AI Spring
Unlike previous AI hype cycles, recent buzz surrounding Generative AI is rooted in real and substantial breakthroughs following a frenzied decade of AI research and advancement. These breakthroughs will undoubtedly change the way we live and work for years to come, rivaling inventions like the personal computer in the scale of their impact.
However, the degree to which that impact will be negative or positive - dystopian or utopian - depends on the way it is put to use. Now is the time for businesses to carefully consider how they can leverage new opportunities arising from Generative AI, while minimizing harmful impacts to their customers and employees.
To help our readers navigate the brave new world of Generative AI effectively, we are beginning a series of blog posts to explore the way this technology will impact many industries, from software development to cybersecurity. To kick things off, in this article we present an overview of Generative AI and a neutral assessment of the hopes and fears surrounding it.
What is Generative AI?
Generative AI refers to a class of AI-driven products which can generate useful data for almost any media type - from text and images to video content - with minimal human guidance. It is distinguished from other AI applications which can recognize and classify data reliably, but not reproduce it.
Generative AI products which have emerged since 2017 – from Stable Diffusion to GPT-4 – are all driven by neural networks, a computing approach which mimics aspects of the human brain. Categories of tools include:
- Text Generation: GPT-4 and its offshoots (ChatGPT, Bing Chat)
- Image Generation: Stable Diffusion, Midjourney and Dall-ECode
- Writing: Github Copilot, codeStarter and CodexVoice
- Synthesis: ElevenLabs, Replica and DescriptChip
- Design: Google, Nvidia and Synopsys
This list is by no means comprehensive: thanks to a continual stream of advancements in the technology behind Generative AI, new use cases and tool categories are being developed at a rapid pace. But just what is that technology, and what makes it so effective?
The Advances Behind Generative AI
While all Generative AI is driven by neural networks, neural networks are not a recent invention – and the concept of AI-powered applications is even older. But until recently, most applications of the technology were underwhelming.
That changed with the development of new machine learning (ML) models in the 2010s and early 2020s. For instance, the arrival of Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) between 2013 and 2014 led to breakthroughs in image synthesis.
Today, these and other advances – including long-short term memory (LSTM) models and transformers – are empowering neural networks to find and replicate patterns in large data sets with little to no supervision, achieving human-like performance for certain tasks.
It is this human-like performance which makes all the difference, compared to AI applications from before the 2010s.
ChatGPT and Large Language Models (LLMs)
Since the arrival of ChatGPT in December of 2022, Generative AI has become almost synonymous with Large Language Models (LLMs) that incorporate neural networks trained on a vast corpus of human text. Recent LLMs created by OpenAI - including GPT 3.5 and GPT 4.0 - utilize transformers to achieve an astonishing range of natural language processing (NLP) capabilities.
While many users see ChatGPT as a magic box, the tool is simply an LLM behind a user-friendly interface, supplemented by a modification which makes the base model responsive to direct instructions (InstructGPT). This combination has made it both a potent tool for productivity, an entertaining toy, and a highly public proof of concept for more specialized Generative AI applications.
LLMs exhibit two features which make them especially useful, and suitable for automating a wide range of tasks:
- Zero-Shot Learning – LLMs can generate multiple content types - such as blog posts, emails and short stories - without being presented with examples, because their training set contained many blog posts, emails and short stories.
- Few-Shot Learning – for more niche tasks – such as test taking, sentiment analysis, or even dialogue generation for video game characters – LLMs can often achieve satisfying performance with only a few examples from the user.
However, the value of LLMs goes far beyond the ability to generate content. Because these models mimic human language, they also mimic aspects of critical thinking and analysis, with the potential to guide granular decision making and analysis. Furthermore, they are not limited to human language – they can write useful code with applications for software engineering, game development, web design, cybersecurity and much more.
General Use Cases
Current and developing use cases for generative AI exist on a spectrum, ranging from assistance for human labor to full automation for complex tasks. No matter where a Generative AI application falls along this spectrum, it can save time, reduce costs, eliminate repetitive tasks, and provide capabilities that organizations could not achieve without them.
- Customer Service and Support – advanced chatbots with access to a database of company or industry-specific information can be deployed through social channels and call centers to provide assistance with a human-like understanding of context.
- Content Creation – LLMs have already demonstrated versatile content creation capabilities, which range from copywriting to creation of webpages, blog posts, eBooks, video scripts, and much more. While human guidance is often needed, LLMs can significantly reduce the time spent on writing rather than research, planning or outlining.
- Media and Entertainment – generative AI applications for music and video creation, voiceovers, storyboarding and auto-translation show great promise as tools for every stage of audiovisual media production. LLMs can also provide on-the-fly dialogue and story content for video games.
- Marketing and Communication – aside from general content creation abilities, LLMs can write personalized messages for customers, press releases, or hundreds of impactful social posts in the blink of an eye. They can also improve on existing marketing collateral, repurposing old content, or updating landing pages to reflect company or product changes.
- Product Development – aside from software development – where Generative AI is already used extensively – emerging Generative AI applications can assist with product design in CAD or chip design in commonly used formats like Veralog. It can also be used to identify vulnerabilities in digital products before they are shipped to customers.
As time goes on, the scope of use cases for Generative AI will certainly expand, since any problem that can be expressed with data is one that Generative AI can potentially solve. Furthermore, AI models can be customized, streamlined and fine-tuned, making them far more efficient for industry and business-specific tasks.
Is Generative AI a Deal with the Devil?
While it’s clear that Generative AI holds great potential as a tool for transforming businesses, it’s understandable that some have reservations. AI has long been depicted as a threatening technology. Besides that, myth and common sense suggest a magical boon never comes without strings attached – and some people think Generative AI comes with a lot of strings.
In April of 2023, the nonprofit Future of Life Initiative published an open letter signed by many AI experts and tech influencers - including Elon Musk - calling for a 6-month moratorium on AI development. In May, “godfather of AI” Geoffrey Hinton left Google, and told the New York Times that “it is hard to see how you can prevent the bad actors from using [AI] for bad things.”
Concerns range from the immediate and realistic, to the remote and speculative. More immediate and realistic concerns include:
- Fake Content – Generative AI can power the spread of disinformation with fake news stories and deepfakes. Some also worry that it will be used to generate spammy messages or fake conversations across social media, turning the Dead Internet theory into a reality.
- Job Displacement – currently, Generative AI is a long way off from automating most jobs. But for knowledge and creative workers – including low-level coders, writers, personal assistants and paralegals – the threat of displacement is real. The threat of integration with robots to displace “blue collar” or manual jobs is not totally remote either.
- Social Engineering – cyber actors often directly contact a company’s employees with the goal of deceiving them into risky actions (such as sharing passwords to sensitive systems or divulging confidential information). Many fear that AI will enable a much larger, more personalized and ultimately more effective volume of social engineering attempts.
- Impact on Students – the existence of LLMs presents a well-established threat in academic environments. Less than a year from the arrival of ChatGPT, up to 43% of college students have already used it to assist with coursework. This may rise to the level of academic fraud, when the AI is used to write entire papers, essays or to solve take-home assignments.
Beyond these, speculative dangers of AI abound, ranging from government takeovers to human extinction. While we won’t address those in this article – beyond noting that they are impossible in the short term and unlikely in the long term – it is worth critically examining the others in the context of tech history.
Technology is Always a Double-Edged Sword
It is a simple and well-established truth that every technology comes with upsides and downsides. In most cases, those upsides and downsides eventually balance out, although it can take society a long time to adjust and may require radically new approaches to old problems.
Examples are easy to come by: cars helped people get around faster, improving tourism, opportunities for social advancement, and national supply chains. But they also gave rise to urban sprawl, pollution, oil dependency and many other issues.
Likewise, the Web and social media have improved global interconnectedness, providing new ways to socialize, do business, communicate and learn. At the same time, they brought about new forms of addiction, an epidemic of misinformation, social isolation, and other problems affecting the world today.
Ultimately there is every reason to believe that – like technologies which have come before it – AI will benefit businesses and consumers in many ways, while also presenting them with new challenges to overcome.
A Cure in the Poison
A recent paper from the Institute for Media Studies in Belgium points out that media often depicts AI as being a “Gate to Heaven” or “Frankenstein’s Monster,” while scientists tend to believe the truth lies in the middle. One reason to have optimism in the face of AI challenges is that AI can help to solve many of the problems it creates.
- AI writing detection is already helping teachers and professors to determine whether their students are submitting AI-generated work; AI also provides learning opportunities that can supplement traditional education.
- Deepfake detection can help Internet users and media organizations to quickly determine if an image has been doctored or fakes and prevent its spread.
- While AI may displace jobs, it will also create new ones, and help workers to thrive in professions where they may have struggled without AI’s help.
For every cybersecurity threat created by AI, AI can also help to find and eliminate vulnerabilities, recognize phishing/social engineering attempts, and educate employees about cyber safety.
Often, those who dwell on AI dangers fail to recognize that AI can – in one way or another – work against those dangers, and level the playing field by giving an equal amount of power to good and bad actors alike. Its cumulative impact on society is a factor of how it is used, not of the technology itself.
A Path Forward
Like any tech advancement with nearly limitless potential, there is no way to put Generative AI back in the bottle and it would be foolish to try. Instead, businesses should be focused on the ways they can use it to better achieve their goals, while improving life for their employees and customers alike.
At the end of the day, there’s no reason that the Brave New World of Generative AI should be a dark one, and early adopters are in a position to make sure that it’s not. Conveyance Marketing Group is here to help: no matter what the future holds, we provide the data-backed strategy and expertise to make the most of new technology and reach your prospects – no matter where they are.