Generative artificial intelligence (or generative AI) exploded on the scene in late 2022, sending people and businesses into a frenzy of curiosity and questions over its potential.

But what exactly is generative AI? Put simply, generative AI is a technology that takes a set of data and uses it to create something new – like poetry, a physics explainer, an email to a client, an image, or new music – when prompted by a human. This marks a shift from traditional AI systems primarily designed to analyse or classify data.

The production of new and original content is why the word generative comes into play.

Unlike traditional AI models, generative AI “doesn’t just classify or predict, but creates content of its own […] and, it does so with a human-like command of language,” explained Salesforce Chief Scientist Silvio Savarese.

Of course, the ability to classify and predict data accurately is a critical element to successful generative AI: The product is only as good as the data it has to work with.

There are several approaches to developing generative AI models, but one that is gaining significant traction is using pre-trained, large-language models (LLMs) to create novel content from text-based prompts. Generative AI is already helping people create everything from resumes and business plans to lines of code and digital art. But the technology’s potential at Salesforce and for enterprise businesses goes beyond making images of polar bears playing bass guitar.

The user gives the tool direction on what to produce, and then, based on the LLMs it has to work with, the AI generates something — be it words, code, or when thinking even bigger, things like novel proteins.

Eventually, Savarese predicts, these AI tools will “assist us in many parts of our lives, taking on the role of superpowered collaborators.” For enterprises, it is especially important to include a human-in-the-loop approach when developing and using generative AI technologies. By doing so, businesses can validate and test automated workflows with human oversight and intervention before unleashing fully autonomous systems. This can help prevent potential risks and ensure that the technology is being used in a responsible and ethical manner. Moreover, having a human in the loop can help build trust and confidence in the technology among stakeholders and customers.

Simply put, researchers feed AI models large training datasets including text, images, audio, or other data types. These models recognise sophisticated patterns and structures within the dataset and can then generate new content based on those patterns.

Given this relationship, the quality and breadth of the training data set significantly impact the AI’s output. For example, an AI trained in a wide variety of human languages can generate more natural and contextually appropriate text.

Several key concepts and technologies (or deep learning models) that enable generative AI tools to function:

1. Neural networks and deep learning
   Neural Networks are the foundational elements of generative AI. These mimic the human brain’s ability to ‘learn’ from data through a process called deep learning. By simulating the way the human brain processes information, they help machines to recognise patterns and relationships within data.
   Deep learning involves multiple layers of neural networks that can learn hierarchical representations. This enables the generation of complex and nuanced content across various data types like text, images, and audio.
2. Transformers
   Transformers are advanced models that help to understand context in human language. This is called natural language processing.
   Many first-time users of tools like ChatGPT are impressed by the advances in transformers, which enable these AI tools to generate coherent and contextually relevant text as responses to human queries at speed. (Note: the GPT in ChatGPT stands for generative pre-trained transformer.)
3. Variational AutoEncoders (VAEs)
   VAEs are like artists who study a bunch of paintings and then create new artworks in the same style. They learn the essence of the data (what makes a Van Gogh recognisable as a Van Gogh, for example) and can generate new items that resemble the originals. The output will be similar but crucially not identical to the original dataset.
4. Generative Adversarial Networks (GANs)
   Imagine two players in a game: a generator and a discriminator.
   The generator creates fake images and new data points based on a dataset.
   The discriminator tries to detect real or fake images by comparing them to authentic data.
   Over time, the creator gets better at making realistic images to ‘fool’ the detector. Through this adversarial training process, GANs work to produce high-quality images and videos: ai-generated content that can potentially pass as genuine.
5. Diffusion models
   These models start with random noise and gradually turn it into meaningful data, like shaping a lump of clay into a sculpture step by step. They refine the data in stages until it becomes a clear and coherent output. Diffusion models have shown exceptional performance in generating high-resolution images and are gaining attention as a viable alternative to GANs.
6. Reinforcement Learning
   Think of training a dog by giving it treats for good behaviour. Similarly, AI models learn the best actions to take by receiving rewards or penalties, helping them make better decisions over time. In this way, AI experts and engineers can shape the desired performance of AI models, fine-tuning them to achieve evermore specific goals.
   For example, AI engineers can use reinforcement learning to reward models for adhering to ethical guidelines when training them.
   
   

Bringing it all together
Generative AI models can perform tasks and create new content across various domains by leveraging these technologies and concepts. The alignment of AI algorithm advancements and greater computational power pushes the boundaries of what machines can create. It’s this relentless progress within the field that is making generative AI such a potentially transformative force.

Generative AI models like ChatGPT, StableDiffusion, and Midjourney have captured the imagination of business leaders around the world.

In fact, a new Salesforce survey found that two-thirds (67%) of IT leaders are prioritising generative AI for their business within the next 18 months, with one-third (33%) claiming it as a top priority.

The technology is “open and extensible – supporting public and private AI models purpose-built for CRM – and trained on trusted, real-time data.”

Salesforce has been exploring how to develop and deploy generative AI to support customer needs for years. For example, the company introduced CodeGen, which democratises software engineering by helping users turn simple English prompts into executable code. Another project, LAVIS (short for LAnguage-VISion), helps make AI language-vision capabilities accessible to a wide audience of researchers and practitioners.

More recently, Salesforce’s ProGen project revealed that by creating language models based around amino acids instead of letters and words, generative AI was able to produce proteins that have not been found in nature, and in many cases, are more functional. With further research, the idea is that these proteins can be used to develop medicines, vaccines, and treatments for diseases.

By making strides in diverse fields like education, virtual reality (VR), augmented reality (AR), and the Internet of Things (IoT), generative AI is creating immersive and interactive experiences.

Ketan Karkhanis, Salesforce’s Executive Vice President and General Manager of Sales Cloud, said that while the technology may be a boon for large businesses, it’s helpful for small- and medium-sized businesses (SMBs) too.

“Capabilities like automated, AI-generated proposals and customer communications, along with predictive sales modeling, will give SMBs even more powerful tools to help them provide great customer experiences, manage operating expenses, and achieve sustainable growth,” Karkhanis said.

Generative AI has the potential to completely reshape the field of customer service. For example, with generative AI layered onto Agenteforce for Service and Einstein 1, businesses have the ability to automatically generate personalised responses for agents to quickly email or message customers, freeing human agents to spend more time deeply engaging on complex issues and building long-term customer relationships.

Generative AI is changing the way businesses operate. Here are a few examples of how different industries and sectors leverage generative AI.

• Enhancing medical diagnosis: Generative models improve the quality of medical images, such as MRI scans, which enhance resolution and reduce noise. This improvement helps doctors make more accurate diagnoses.
• Improving drug discovery: AI models can generate potential molecular structures for new drugs, accelerating the initial discovery process. This is due to their ability to analyse vast amounts of data to propose new compounds that may be effective against specific diseases. 

• Maximise efficiencies: AI can help relieve the burden on healthcare professionals by assisting with some administrative tasks. 87% of healthcare leaders believe AI can help reduce burnout issues. Better imaging and AI-driven, personalised treatment plans lead to better healthcare overall. Doctors recognise AI as a tool they can leverage to boost their service while still being a people-centred profession.

Generative AI also reduces the time and resources required to develop treatments and manage patient care. These efficiencies translate into significant long-term cost savings for healthcare providers and patients.

Yet, even in healthcare, there are drawbacks. First, there is still a question of reliability. We must thoroughly validate AI models to ensure they provide accurate and safe recommendations. Even though humans are fallible, many of us would still prefer to rely on human expertise over an algorithm when it comes to health outcomes.

There’s also a data privacy issue. Handling sensitive patient data requires strict compliance with privacy regulations.

Customers across India are increasingly expecting quick, efficient, personalised customer service interactions. AI is helping businesses meet these expectations, with almost 80% of service professionals who use AI saying it saves them time.

• AI-powered chatbots like ChatGPT can handle customer questions and offer personalised recommendations. These can feel like ‘real’ conversations, thanks to the latest advancements.
• AI can generate email replies based on customer queries, making communication more efficient for all stakeholders.

AI agents don’t need breaks; they can simultaneously provide 24/7 support to many customers. When done well, this improves customer satisfaction. Since many companies have access to customer data, AI can offer tailored solutions. Even though you’re talking to a ‘robot,’ the interaction feels more real thanks to the hyper-personalisation made possible through access to data.

How Salesforce Is Building the Business AI Agents of the Future | Salesforce

Yet, these customer interactions aren’t perfect. AI may not fully grasp the emotional nuances of human interaction. Since customer service is based on trust, this lack of perceived empathy can be damaging. AI is also capable of making other mistakes. Errors in understanding, whether through poorly phrased input or a misunderstanding of context, can lead to inappropriate or simply factually incorrect statements.

Companies in the finance sector are in a race to develop tools that use generative AI to predict market trends. They’re also using it to improve access to personalised advice.

• AI models generate trading strategies by analysing historical data, aiming to optimise investment portfolios.
• Generative models simulate various financial scenarios to help institutions assess potential risks (and prepare accordingly).
• AI chatbots can offer financial advice to customers, boosting access to financial advice.

AI can provide insights that lead to better financial strategies, both for corporations and individual investors and savers. With incredible computing power, AI models can complete calculations and simulations more efficiently and accurately than humans can imagine.

Yet, markets are unpredictable. Over-reliance on AI models may be risky due to unforeseen events. For instance, how can these models predict another pandemic that results in a black swan event? Another drawback is that these financial AI applications are currently experiencing unclear regulatory issues.

While the potential of generative AI is enormous, it “is not without risks,” according to Paula Goldman, Salesforce Chief Ethical and Humane Use Officer, and Kathy Baxter, Principal Architect for Salesforce’s Ethical AI practice. In a co-authored article, the pair pointed out that it’s “not enough to deliver the technological capabilities of generative AI. We must prioritise responsible innovation to help guide how this transformative technology can and should be used — and ensure that our employees, partners, and customers have the tools they need to develop and use these technologies safely, accurately, and ethically.”

Building trustworthy generative AI requires a firm foundation at the inception of AI development. Earlier this year, Salesforce published an overview of our five guidelines for the ethical development of generative AI that builds on our Trusted AI Principles and AI Acceptable Use Policy . The guidelines focus on accuracy, safety, transparency, empowerment, and sustainability — helping Salesforce AI engineers create ethical generative AI right from the start.

Beyond regulation, part of the solution will be to develop AI tools that can detect and flag AI-generated content reliably.

In an interview with Silicon, Goldman shared, “Accuracy is the most important thing when applying AI in a business context because you have to make sure that if the AI is making a recommendation for a prompt, for a customer chat or a sales-focused email, that it’s not making up facts.”

The authoritative feel of ChatGPT responses is itself something to be mindful of, said Savarese, who warned it could lead to what he deems “confident failure.”

“The poised, often professional tone these models exude when answering questions and fulfilling prompts make their hits all the more impressive, but it makes their misses downright dangerous,” Savarese said. “Even experts are routinely caught off guard by their powers of persuasion. Scale the reliance on tools like ChatGPT up to the enterprise level and it’s easy to see how high the stakes could get. But IT leaders are on guard. Nearly six in 10 (59%) said they think generative AI outputs are inaccurate.

Then there’s the question of how to use generative AI ethically, inclusively, and responsibly.

That’s why Salesforce is building trusted AI capabilities with embedded guardrails and guidance to help catch potential problems before they happen. If the world is going to realise the potential of generative AI, it will need good reasons to trust these models at every level.

While there are specific limitations within sectors, listed below are a few broader challenges:

Bias in AI Models:

Generative AI creates new content based on its training data. As such, the new content reflects the training data. So, what if the data is flawed in some way? Logic follows that the output will also be flawed.

Of course, AI professionals see the solution in better AI. Implementing responsible AI practices, such as bias detection algorithms and diverse training datasets, can mitigate this issue.

Ethical and Secure usage of AI:
Ensuring data is accurate and trustworthy is foundational to any AI application. This is where the alarming rise of deepfakes and disinformation is concerning. Generative AI has lowered the barriers of entry for those looking to sow discontent by the spread of deliberately malicious content and this poses risks to our society.

But sometimes, AI can be harmful even without malicious intent. For example, what would happen if a security camera used facial expressions to identify a scenario as ‘safe,’ but in this case, it was not? Who will be responsible for this?

Determining responsibility for these AI actions is complex. When AI systems make mistakes, it's unclear who is accountable — the AI developers, users, or the AI itself.

The takeaway here is that we need whole new legal frameworks to define accountability as it relates to AI.

Governments are working hard to reign in the AI wild west. It’s often the case that technological advances outpace government reactions. Regulation is needed worldwide to establish ethical guidelines and to curb some of the more harmful AI use.

Additionally, promoting digital literacy will also help the public identify and question suspicious content. With AI's ever-growing prominence on our screens, digital and media literacy is likely to be needed to be taught in schools.

Computational Costs:
It is hard to ignore the computational costs of using generative AI. Training and running advanced AI models require significant resources. ChatGPT costs more than $100,000 a day ( approx. INR 8.6 million) to run alone. However, developing more efficient algorithms, utilising cloud computing resources, and potentially accelerating investment in quantum computing research may reduce some of these demands.

Data Privacy:
Widespread integration of AI also raises the challenge of maintaining data privacy. Collecting and processing large amounts of personal data can infringe on individual privacy rights. In fact, some of the best use cases for generative AI fundamentally rely on data collection. The solution for companies is to comply fully with European privacy laws, such as GDPR.

Further, to gain public trust, companies should ensure transparency regarding data management practices. In some cases, customers can opt in or out of data collection, understanding that they know how their data can be used by opting in.

Job Displacements:
Automation through AI may lead to job losses in some sectors. Workers in roles susceptible to automation might face unemployment or the need to reskill, especially if they don’t have protections in place in the form of unions. There’s no easy solution to the risk of job displacement due to AI, and your response to the issue likely depends on your philosophical viewpoint on the role work plays within society.

On a more positive note, generative AI will, in fact, also give rise to a myriad of new jobs for human beings. For example, high-paying roles like prompt engineer — essentially, a master of the art of crafting prompts for GPT interfaces — and AI product manager are currently trending on popular job search sites.

According to the Salesforce-sponsored IDC white paper of 500 organisations using AI-powered solutions, companies will also see a sharp increase in hiring for data architects, AI ethicists, and AI solutions architects in the near future. That same report predicts 11.6 million new jobs will be created within the Salesforce ecosystem alone over the next six years.

Infosys, a global consulting and IT services leader utilises generative AI to improve software development and business solutions.

Infosys uses AI-powered tools to generate code snippets, suggest optimisations, and automate specific coding tasks. They examine existing codebases and leverage machine learning models to understand complex coding patterns and best practices.

Leading AI and cloud computing innovation: Infosys' strategic use of GitHub

This is a significant upgrade for developers who can now use AI as code assistance to accelerate the development cycle. If a team can reduce the time spent on some of these routine coding tasks, they can focus more on problem-solving and innovation. The result is increased productivity overall, all thanks to AI’s ability to lighten the load.

We’re only human. Even coders make mistakes. Automated code reviews powered by AI help detect errors and improve software reliability. The AI models can identify bugs, security vulnerabilities, and deviations from coding standards early in development. This proactive approach leads to higher-quality software and reduces the cost and effort of fixing issues later in the life cycle.

For Infosys, this means they provide better solutions to their clients, pure and simple.

Of course, it would be important for us to mention the challenges. Developers need to become familiar with the new tools and understand how to incorporate them into their daily work. This learning curve can temporarily affect productivity. They’ll also need training, which can be costly.

Another concern is the potential dependence on AI-generated code. More reliance on AI assistance might impact developers' skill development and understanding of underlying programming principles.

In short, developers could become, well, lazy. Infosys looks to mitigate this by promoting an approach whereby AI serves as a complementary tool to humans rather than something intended to replace them.

The company encourages continuous learning and critical thinking, ensuring developers review and understand the AI-generated code before implementation.

Generative AI is becoming increasingly accessible, with various AI platforms and tools available for individuals and businesses. Here are some popular options to explore:

• ChatGPT by OpenAI: An AI language model that can generate human-like text. It can help draft emails, write code, do maths or even compose AI content like poetry. It's a prime example of how transformers enable natural language processing.
• DALL·E 3 image generator: Also from OpenAI, this tool generates images from textual descriptions. It's now integrated into ChatGPT. Artists and designers use DALL·E 3 to bring creative concepts to life quickly and easily.
• Google Gemini: A conversational Google AI service that assists with research, drafting content, and answering questions across a wide variety of topics.
• Salesforce Einstein AI: Integrated within Salesforce's CRM platform, Einstein uses AI and machine learning to deliver predictions and recommendations. Businesses can leverage Einstein AI to enhance customer experiences, automate tasks, and make data-driven decisions.

If this has piqued your curiosity, mentioned below are pointers that can help you deepen your understanding of generative AI solutions:

• Consider AI and machine learning courses on Coursera.
• Check out the free resources on Trailhead and learn at your own pace.
• Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig is a great book offering a comprehensive guide to AI principles and practices.
• If you’re super keen, dive into the latest research papers on transformer models and GANs.

AI Stack Exchange and GitHub are two online communities where you can browse forums and access repositories related to AI.

The potential of generative AI is vast and continually expanding. Here are some of our predictions for where the technology may lead:

• There will be even more personalisation in areas like education, healthcare, and marketing. AI systems can generate content and recommendations that meet specific needs based on more data analysis.
• AI models are expected to become even better at understanding and generating realistic human language. This will make interactions with virtual assistants and chatbots more natural.
• Combining generative AI with virtual reality (VR), augmented reality (AR), and the Internet of Things (IoT) could create immersive and interactive experiences. For instance, AI-generated environments in VR could adapt in real-time to user inputs.
• Policymakers and industry leaders will need to collaborate to establish guidelines and regulations as generative AI becomes more prevalent.
• Companies will rise to meet the challenge of computational and environmental costs, developing more sustainable practices.

Emphasis on reskilling and education will be crucial to prepare the workforce to work alongside generative AI.

AI has long been integral to the Salesforce Platform. For example, Einstein AI technologies deliver over 200 billion daily, helping businesses close deals faster, provide AI-powered human-like conversations for frequently asked questions, and understand customer behaviour better.

Einstein by Salesforce is the world’s first generative AI for CRM. From personalised sales emails to auto-generated code, Einstein delivers AI-created content across sales, service, marketing, commerce, and IT interaction, at hyperscale. And, it’s built for customers in a way that’s relevant to them.

Einstein uses data from Data Cloud combined with public data to create content. And, it will do so with the same foundation of inclusivity, responsibility, and sustainability that is at the core of any Salesforce product.

Read about generative CRM and what it means for businesses.

Learn more about Salesforce Einstein AI Solutions and how it marks the next big milestone in your AI journey.