Walter Writes vs WriteHuman

Generative AI is going to completely alter (restructure) knowledge work. While it will make many aspects of our jobs easier (and potentially less time consuming), it will also create the opportunity for us to be judged for our decision making and judgment.

These are just some examples of things that AI does well:

generating boilerplate code drafting emails summarizing documents creating initial drafts of most types of written content

All of these tasks used to take up a significant amount of a worker's time. Research done at MIT showed that employees who use ChatGPT to assist with writing tasks finish their assignments about 40% faster and have higher quality ratings. Similar results have been seen from research done at GitHub regarding developers and the task completion times with and without the assistance of Copilot. The closer we get to a point where generating a first draft costs nothing, the less valuable the first draft becomes. What takes on greater importance is all the other steps involved in creating a final product. Who determines if there is enough justification to pursue a particular issue? Does the user recognize when a system is providing a false confidence in its ability to provide an answer?

One of the problems with generative AI is that it often generates information based upon the assumption that the user wants "right" answers. This means it sometimes provides false citations, misreads contextual clues provided in the question, and/or makes flattering assumptions based upon the input provided. In addition, a lawyer who provides a court citation produced by AI in an effort to support his argument, such as was demonstrated in the recent *Mata v. Avianca* court case, has not lost their job due to the introduction of AI into their profession. Rather, the lawyer failed to utilize their analytical skills when reviewing their potential argument before presenting it to a judge. Verifying accuracy and being accountable for one's actions remains the responsibility of humans. As the volume of generated material increases, so too does this responsibility.

There is a very real concern related to this trend, particularly concerning entry level professionals. Traditionally, entry-level positions within a field have acted as an educational experience for young professionals. Junior analysts learn by drafting memos and junior developers learn by identifying and correcting errors in programs. With AI absorbing much of this type of work, companies will need to identify new methods to train professionals or else they will lose the individuals who know how to supervise these systems. Companies that view AI as solely a method to cut costs may end up eliminating their own talent pipeline.

Another criticism is that improvements in capability are happening rapidly enough that even areas that require high levels of judgment will be able to be automated. This is certainly possible. However, historical trends suggest that while automation replaces many of the rote, repetitive tasks associated with a position, it rarely eliminates an entire occupation. Spreadsheets did not replace accountants; instead they reduced the number of hours spent performing arithmetic and increased what accountants needed to analyze.

As always, the best way to respond to innovation is not with either panic or complacency. Professionals should focus on developing their abilities in problem identification and analysis; critical evaluation; and area-specific expertise. Organizations should design new roles and training opportunities focused on human oversight of AI.

# Generative AI and Knowledge Work

Typing a description of generative AI into a search engine yields two major camps of thought. The first camp believes the widespread adoption of generative AI will result in large scale automation of analysts, writers and programmers. The second camp believes these technologies are little more than glorified autocomplete systems and therefore will have little impact on the workforce.

The reality will be more nuanced. Routine cognition will be commoditized and the value of ethical judgment and accountability will increase.

Generative AI is effective at a myriad of cognitive tasks. These systems can now perform many activities that comprise a knowledge worker’s job. For example, Email, document and code drafting. First iteration generative systems have shown their capabilities and dramatically reduced the time it takes to perform writing and coding tasks. Researchers at MIT and the famed EIC (Economic Intelligence Unit) at Forbes found that ChatGPT increased the quality of writing and reduced the time required to write a document by 40%. Likewise, developers using Copilot dramatically increased their productivity and reduced the time it takes to perform software development tasks.

What gets valuable are the things adjacent to the draft. This includes deciding what problem to solve, assessing the fidelity of the output, understanding when a seemingly correct answer is actually false, and recognizing flaws in others’ thinking. There have been major concerns about the quality of Large Language Models and the confidence with which they make assertions, and these issues are becoming more amplified. These models may make users of the technology feel as though they’ve been replaced, but in truth, these users have often simply failed to adequately evaluate the models’ outputs and exercised poor professional judgment. Verifying the integrity of content and holding parties accountable are activities that will become increasingly important and that, for the foreseeable future, will require human interaction.

While technologies to assist knowledge work are not new, automation at this scale to perform tasks that require critical judgment and professional skills, such as drafting memos and analyzing data or legal frameworks, would dramatically reshape the practice of knowledge work and potentially displace human professionals. However, there is an opportunity for companies to redesign how they integrate and deploy this technology to address the needs of their business while preserving professional personnel and mitigating risks associated with technology displacing human professionals.

Critics believe that automation will eventually impact even the most judgment-based jobs. While it’s true that virtually no job will be completely immune to automation, it’s important to recognize the net positive impacts that automation has historically had on the labor force. The introduction of spreadsheet technology, for example, ultimately enabled accountants to spend less time on mundane, repetitive tasks and led to a greater focus on higher-level analysis.

Automation and the changing nature of work present both challenges and opportunities to the workforce and to the employer. The nature of jobs should be redesigned by employers to incorporate relevant oversight by a human operator and to give employees opportunities to leverage and deepen their critical analysis and domain-based skills. The job of generating ideas and determining what to do with the new ideas brought on by Generative AI will be a greater challenge for the workforce than figuring out how to incorporate this technology in the workplace.