Let’s use five ways to apply responsible generative AI techniques at work

Midjourney, ChatGPT, Bing AI Chat, and other AI tools make generative AI easier to use, and these tools create a wealth of ideas, experiments, and creativity. If you want to use generative AI in your organization, you're still faced with the question of where to start making AI work, and how to do it without running into ethical dilemmas, copyright infringement, or factual errors. A good start is to use generative AI to empower people who are already experts in their field, helping them save time and become more productive.

There are many other ways you can start using generative AI right away, and generative AI may already be integrated into several tools and platforms your organization is using. Therefore, you need to consider developing a guide on how to trial and adopt these tools. Here are five key areas to consider using generative AI, along with guidance on finding other suitable scenarios.

1. Improve developer productivity skills

People often think that coding is somewhere between art and science, but a lot of work in programming is routine and repetitive. The rise of cloud platforms and module repositories means that writing modern applications and bringing together components and APIs, refactoring existing code, optimizing environments, and orchestrating pipelines is as important as coming up with algorithms. Much of this work is ripe for automation and AI assistance, but equally, you need to know how and where to use these tools to monitor their impact and effectiveness. Before you make the full switch to coding assistants, you can start with one-off tools that speed up specific common tasks.

Documentation is both important and often overlooked: not only can you let generative AI document the code base, but you can also build a chat interface into your documentation that allows developers to ask questions through the interface about how it works and how to use it. way, or just replace the usual search box, turning generic documents into conversational programming, where the AI ​​can, for example, take the data and show how it wrote the query.

Testing is another area that is easily overlooked, and automatically generating unit tests can help you expand your testing scope. Submission bots can also help developers write messages that contain enough information to help users and other developers, while generative AI can do the same for IT staff who log upgrades and system reboots.

It’s also critical to tell the AI ​​what you want to generate backend logic and other boilerplate so developers can focus on the more interesting and creative parts of the app. You should also leverage generative AI to write your own code modules (scripts that automate repetitive, time-consuming tasks in large code bases) or leverage generative AI to help fix the voice and tone to better fit in-house styles. Coding assistants like GitHub Copilot and IDEs built into large language models (LLMs) can do all of this and more, but shouldn't replace developers; these coding assistants and IDEs need to understand and evaluate code that is not written ( and the context in which it runs) in case it contains security vulnerabilities or performance bottlenecks, omissions, bad decisions, or just plain mistakes, since it is learning from repos that may contain any or all of the issues to generate code . You’ll want to consider how AI-generated code will be tracked across your organization so you can audit it and evaluate its usefulness. Developers report becoming more efficient and reducing frustration using GitHub Copilot. Microsoft says that 40% of Copilot users' code is AI-generated and unmodified. Currently, this provenance is lost once a developer leaves their IDE session, so consider documenting internal guidelines for how AI tools are used.

2. Improve the skills of low-code and no-code business users

Although business users do not have the expertise to evaluate the code generated by AI assistants, low-code and no-code environments are highly constrained and Where generative AI tools are integrated, problems are much less likely to arise.

Low-code applications require frequent retrieval and filtering of data. Low-code platforms have added generative AI capabilities that can generate lookup queries or clean returned data — such as programmatically adding missing zip codes — which allows business users without database expertise to go one step further without having to persist. Use pre-built components or wait for professional developers to write query strings for them. Open source tools like Census GPT make it easier to query large public datasets.

Code Assistant isn’t just for professional developers, either. Wix Artificial Design Intelligence (ADI) can build you a complete website, combining code generation and generative design; Uizard can do the same for website and app prototypes; Fronty can convert images to HTML and CSS, and Microsoft Power The Express design function in Apps can convert hand-drawn sketches or Figma files into runnable applications, which is completed on the backend.

Most generative AI uses of interest to enterprise organizations are modules that can be called within low-code automated workflows so that employees can tailor them to their specific needs. Moreover, many platforms already provide ChatGPT and other OpenAI APIs like any other component. However, make sure any warnings or guidance that comes with the generated text or images are displayed correctly in a low-code environment, ideally there is a way to provide feedback, and staff know your policy if this can be shown directly to customers without requiring staff to advance Review these.

3. Understand documents and data

Combining the customized version of ChatGPT with Bing has brought millions of new users to Microsoft's Bing search engine. But the way LLM works means errors and "illusions" can occur because they essentially autocomplete sentences and paragraphs to produce text that matches the query prompt. If the information you want does not exist, the model will still try to create some reasonable information, and even if the information provided is correct and consistent with what most experts in a field say, the answer may be incomplete and incomplete. Accurate, and if you're not an expert yet, you probably don't even know what you're missing. These issues are a big problem for both enterprise search and the public web; Microsoft's upcoming Microsoft 365 Copilot tool will try to solve this problem by querying and providing references to Microsoft Graph data based on documents and entities, but some may still be missed Key points need to be added by yourself.

Start looking for opportunities to leverage LLM to summarize and analyze documents, or generative text to explain concepts in those more restricted scenarios where information is reviewed internally by people with expertise , rather than being displayed directly to your customers or other end users.

Generate a knowledge graph to visually display the connections and relationships between different entities to help you understand the project, community or ecosystem. The Copilot tool in Excel enables you to gain insights in an interactive way and ask questions about your data in a sandbox that does not change the underlying data, so any mistakes may lead you down the wrong path but should not contaminate Raw information for future analysis.

Telling stories with data is another effective way to communicate key trends and AI analysis. Smart Narratives in Power BI, for example, can find anomalies and influencing factors and then describe them with charts and automatically generated descriptions. explain. This avoids the mathematical problems faced by LLM, as insights are derived by AI models such as linear regression and then described by language models. This integrated approach may become more common. Similarly, security tools are starting to use language generation to explain threats, anomalies, and evidence of potential breaches detected by AI, telling you in clear and personalized language what it means and how to respond. In the future, the hope is to be able to ask questions to these tools and have them interpret the suggestions they give.

You can also make existing chatbots smarter and more flexible, from keywords and fixed response content to making response content sound more natural and automatically included when the knowledge base is updated. new information. Likewise, it is tempting to use generative AI to interact directly with customers to improve customer satisfaction and reduce costs, but this is more risky than using generative AI within an organization to surface useful information about benefits and other HR issues. Bigger. While trendy chatbots are suitable for some brands, you don’t want to make headlines because a customer received dangerous advice or was insulted by your chatbot. Using generative AI to provide agent assistance allows you to increase productivity while reducing risk.

4. Accelerate the workflow of business users

Meetings should be the place where business decisions are made and knowledge is shared, but the value of a meeting never leaves the conference room. AI tools like Microsoft Teams Premium, Dynamics 365 Copilot, and the ChatGPT app for Slack can generate summaries and record assigned task entries for attendees and those who are not in the room and may not know what they are responsible for, which can also help avoid questions about The tug-of-war over who is asked to take notes and who does other “routine office tasks.”

Being able to catch up with the busy pace of Slack every day can also improve productivity and work-life balance, but the people making plans and decisions should be responsible for ensuring the accuracy of AI-generated summaries, action items, and timelines. AI tools that summarize the content of phone calls with customers can help managers supervise and train employees. This may be useful for financial advisors and call center workers, but tools to monitor employee productivity need to be used in an empathetic way and avoid concerns raised by workplace surveillance. User feedback and product reviews are helpful, but the sheer volume of information can be overwhelming, and useful information can be hidden deep within.

Generative AI can classify, summarize, and categorize corresponding content to provide aggregated feedback that is easier to absorb. In the long term, it's easy to imagine having a personal shopping assistant make recommendations on what items you want to buy and answer questions about those items, rather than leaving you to scroll through pages of reviews yourself. But equally, companies must be wary of introducing tools that might elicit offensive or defamatory opinions, or that are overzealous in filtering out negative reactions. Generative AI tools can read and summarize long documents and use the information to draft new documents. There are already tools like Docugami that can extract due dates and deliverables from contracts, and international law firm Allen & Overy is trialling a platform to help with contract analysis and regulatory compliance. Generating semi-structured documents such as memoranda of understanding, contracts, or statements of work may speed up business processes and help you standardize some business terminology programmatically, but expect this process to require a lot of flexibility and oversight.

5. Overcome writer's block and beautify your design

You don't have to hand over the entire writing process to AI to help you brainstorm ideas, write copy, create images or designs. Soon you'll be able to ask generative AI to create documents, emails, and slides through Office 365 and Google Docs, so you'll need to have policies in place about how to check the accuracy of this content before sharing it with anyone. Likewise, you should start with more limited tasks and internal uses that can be monitored.

Generative AI can suggest what to write in customer outreach emails, thank you notes, logistics issue warnings, right in your email or in a CRM like Salesforce, Zoho or Dynamics 365 as a platform part or implemented through third-party tools. There is also great interest in using AI for marketing, but there are also brand risks. You should think of these options simply as a way to get started, rather than a final version before clicking send.

AI-generated text may not be perfect, but if you have a lot of gaps to fill, it's better than nothing. For example, Shopify Magic can take detailed basic information about a product and write consistent, SEO-tuned product descriptions for your online storefront, and once you have some content, you can improve it. Additionally, Reddit and LinkedIn use Azure Vision Services to create captions and alt text for images to improve accessibility when users don't add this content themselves. If you have a large library of training videos, automatically generated summaries may help employees make the most of their time. Generating images from text is incredibly powerful, and tools like the Microsoft Designer app can put the image propagation model into the hands of business users, who may be reluctant to use a Discord server to access Midjourney and don't have the expertise to use Photoshop. Stable Diffusion plug-in. But there are also controversies surrounding AI-generated images, from deepfakes and the uncanny valley effect to the origins of training data and the ethics of using the work of famous artists for free. Organizations will want to have a very clear policy regarding the use of generated images to avoid obvious pitfalls.

Find Your Purpose

As you can see, from customer support and retail to logistics and legal services, in any interaction you want to leverage a reliable source of information for curation, All have opportunities to benefit from generative AI.

To use generative AI responsibly, start with natural language processing, such as classification, summarization, and text generation for non-customer-facing scenarios where the output content needs to be produced by experts with discovery and Review by people with expertise in correcting errors and disinformation, and have an interface that makes this process easier and more natural than just accepting suggestions. It’s tempting to save time and money by skipping human involvement, but if the content generated is inaccurate, irresponsible, or offensive, the damage to your business can be significant.

Many organizations worry about leaking data into models that could help competitors. Google, Microsoft, and OpenAI have all published data usage policies and said that the data and hints used by enterprises will only be used to train their models, not in the core models provided to each customer. But you still have a guideline on what information employees can copy into public generative AI tools.

The manufacturer also stated that users have ownership of model inputs and outputs, which is a good idea in theory, but may not reflect the complexity of generative AI when it comes to copyright and plagiarism issues, and models like ChatGPT Quotes are not included, so you don’t know whether the textual content returned by the generative AI is correct or copied from someone else. Paraphrasing isn’t exactly plagiarism, but stealing someone else’s original ideas or insights is never a good thing for any business.

It is also important for organizations to develop AI literacy and familiarize employees with using and evaluating the output of generative AI. Remember, you have to start small in the areas that don’t matter and learn from the areas that work.

The above is the detailed content of Let’s use five ways to apply responsible generative AI techniques at work. For more information, please follow other related articles on the PHP Chinese website!