What the Big Tech Companies Won’t Tell You About AI

AI is all over the news. It’s all over social media. Talking heads are debating whether it will take over your job or make you so productive you’ll actually be able to take that long-postponed vacation.

But most of the information we are exposed to from its creators neatly gloss over the realities AI brings to the table. What are they not telling you?

For one thing, it’s not a silver bullet for all ills. AI is constantly evolving, and while it can do some genuinely amazing things right now (like pass the bar exam or beat chess masters at their own game) its capabilities overall are overhyped right now.



What’s the Catch?

Trust and reliability issues abound with generative AI. (Remember the lawyer who used ChatGPT to write a brief that included fabricated cases? Oopsie.) It’s critical to remember that a human needs to remain in the loop for the foreseeable future to prevent situations like this from occurring.

One-size-fits-all AI is a recipe for disaster. Industry-specific AI tools, tailored to legal practices for accuracy and data protection, will be the key to adoption and success.

There is a general unease with AI. Humans are naturally suspicious of things they can’t inherently explain, and AI is no different.



Generative AI Has Trust and Reliability Issues

Still in its nascency, generative AI products that operate without the proper safeguards can produce inaccurate information or "hallucinations" (gaps in data that are filled with fabricated content). Simply put, if generative AI doesn’t have the data it needs, it will create it. And you have no control over what it decides to come up with. That’s a significant problem to overcome, but achievable by putting parameters in place for the algorithm to adhere to.



Ensuring Accuracy, Reliability, and Transparency in AI-Generated Outputs

By extracting information into key-value pairs to classify and manage data effectively within legal workflows, users can verify the consistency and accuracy of information generated by AI tools, ensuring that the data input and output align correctly.



What is a Key Value Pair?

In generative AI, a key-value pair is like a matching set. Think of it as having a label (the key) and what it represents (the value). An example of a key-value pair in the context of generative AI could be an ICD code paired with its corresponding code itself.

For instance, an ICD code for a specific medical condition could be paired with the actual code representing that condition. By ensuring that the key (ICD code) is linked correctly to its corresponding value (actual code), users can verify the consistency and accuracy of information generated by AI tools, particularly in tasks like drafting documentation or managing medical records.

This structured approach allows for precise data storage, retrieval, and processing, enabling users to track and validate the information produced by generative AI systems. Understanding key-value pairs in generative AI helps users maintain control over the data inputs and outputs, enhancing trust and reliability in AI technologies and optimizing their application in specific tasks and workflows.



Generic AI Tools Don’t Inspire Trust

Industry-specific tools provide a construct for prompts that ensure only certain pieces of data are generated, unlike generalist platforms like Chat GPT, which may lack specificity. Furthermore, understanding where the data comes from and how AI outputs information is critical to maintaining transparency and reliability. If the AI model is only given access to accurate datasets verified by the user, fabrication is no longer an outcome.

By demystifying the inner workings of complex AI systems and providing transparency on data sources, users can verify the accuracy and reliability of AI-generated outputs. Big Tech doesn’t want users to question where their data comes from, but that’s exactly what you need to do to build trust and ensure the proper use of AI tools within legal practices.



Shhhh: Don’t Tell, but AI Still Needs Humans!

There is a significant need for caution in relying on technology as infallible. (See the earlier example of the enterprising lawyer if you have any doubts about this one.) Adopting a human-in-the-loop approach will ensure accuracy and reliability in legal workflows.

Keeping the human in the loop is crucial in AI implementation for several reasons. One key aspect is the need for human oversight to review AI-generated outputs for accuracy and reliability, especially in tasks like drafting documents or demand letters. Despite the efficiency gains from AI tools, human verification ensures that the information produced is correct and aligns with the desired outcomes. By involving humans in the review process, organizations can mitigate errors and discrepancies in AI-generated outputs, enhancing the overall quality of work.

Additionally, maintaining human involvement fosters trust and transparency in AI technologies, as employees can understand and validate the outputs, leading to increased confidence in utilizing AI tools effectively. The human-in-the-loop approach not only improves the accuracy of AI-generated results but also empowers employees to leverage AI as a supportive tool in their workflows, driving efficiency and productivity while ensuring data integrity and reliability.



Concerns around Data and Privacy

Every day we are bombarded by one story after another about hackers and data leaks. The risk of security breaches when using some generative AI is very real, especially when sensitive information is copied and pasted into platforms like Chat GPT, potentially leading to data exposure. Unknowingly exposing critical information to AI models poses a significant security risk for organizations.

Additionally, the storage and sharing of personal and business data by big tech companies raises privacy concerns, as data may be shared with affiliates and other entities without explicit consent. Understanding how data is used and stored by technology providers is crucial to safeguarding sensitive information and mitigating the risk of data exfiltration.

Implementing a zero data retention policy (like NeosAI has) where data is only used for immediate tasks and then scrubbed and cleaned, can help protect against data breaches and unauthorized access. Educating staff on acceptable data use, avoiding free platforms for sensitive information input, and ensuring data rights and privacy protection are essential steps to address data and privacy concerns in AI adoption within legal practices.



Employee Adoption as a Barrier in AI Implementation

Big Tech doesn’t want to admit that not everyone is enamored with their products. Employee adoption poses a significant barrier in AI implementation.

Fear of job loss among employees, stemming from misconceptions that AI will replace human roles, is rife. This fear can lead to resistance in adopting AI technologies and hinder employees from embracing AI tools as a means to enhance their productivity and skills.

Additionally, the complexity of AI technology and terminology can overwhelm employees, making it difficult for them to understand and utilize AI effectively. Lack of education and training on AI tools further contributes to low adoption rates, as employees may feel ill-equipped to incorporate AI into their workflows.

Encouraging employee education, providing training, and promoting a human-in-the-loop approach can help address these adoption barriers and facilitate successful integration of AI technologies in the workplace.



Strategies for AI Adoption

Embracing a trial-and-error approach within a controlled environment to learn and build confidence in using AI effectively and safely is a savvy strategy for adopting AI. By encouraging staff to experiment with AI tools, provide feedback, and ensure human oversight (always reinforcing the concept of "human in the loop"), organizations can maximize efficiency in initial drafts and summaries while maintaining accuracy through human verification.

Additionally, individuals should seek industry-specific AI tools tailored to address specific workflows and problems, rather than relying on generalist platforms, to ensure that AI solutions align with their unique needs and constraints.

Lastly, promoting job resilience and viewing AI utilization as a skill for future growth can empower employees to embrace AI tools, experiment with different technologies, and adapt to evolving workflows, ultimately driving efficiency and productivity within the organization.

While Big Tech may casually dismiss any concerns that don’t fit their prescribed narrative of AI being a revolutionary disruptor, nothing comes without strings. Therefore, asking questions and understanding how to use AI effectively is the best way to enjoy the benefits and sidestep the risks.

Watch the full webinar on What Big Tech Doesn’t Tell You About AI, hosted by McKay Ferrell, SVP of Product at Assembly.