Google Addresses Negative Criticism Surrounding Its AI Overviews

The launch of AI Overviews for search engine giant Google has not gone as planned.

The company has faced global criticism for introducing the feature across its Search and now, the search giant is finally breaking its silence on this front.

The company’s head for search was quick to explain through a new blog post how the feature really works and how the idea arose, to begin with. This includes what steps the firm is taking to ensure user safety and amend any errors that have been in discussion since the launch.

But it was interesting to witness the company add how plenty of searchers across Google were getting greater satisfaction across results and this has allowed them to throw more complex scenarios and queries in Google’s direction, hoping to find a reliable and accurate result.

So if you think AI Overviews are going anywhere, well, that’s certainly not the case, unfortunately.

Google mentioned how the rollout is very effective and has been in the works after a careful testing period. If that was not enough, it promised stats on how the accuracy rates for such features are at the same level as that seen for snippets. The latter also makes use of AI.

Coming down to the topic of hallucinations and if the features do embark on them, the answer was no from Google’s end. Many have been criticizing the feature for making up lies that don’t exist but the Android maker says that’s not the case.

The mistake only arises if the search engine misinterprets questions doesn’t comprehend the right language online or doesn’t have great sources for information available.

Such odd results still keep a lot of people up at night but Google says it doesn’t understand why so much hate is being generated. It mentioned how internal testing proved fabulous results and extensive work and research was done before the rollout, and only then did they feel comfortable going ahead with the release.

Google even admitted how plenty of searches were being done intentionally to prove the feature wrong and those which didn’t even make any sense were put forward, leading to inaccurate results and false hype being generated in that context.

Faking examples of instances is also on the rise, Google warned that people were manipulating screenshots displaying false AI replies where AI Overviews were never generated.

Some very odd examples popped up, Google admitted and it hopes to make more improvements on this front. It doesn’t intend to make manual adjustments to AI Overviews but hopes to enhance models so they can encompass more queries than before.

This can only be done through updates, Google confirmed, to give users the best search experience that they are entitled to.

There was also a discussion on data voids in the latest blog post where Google says it’s covered that aspect several times before so it was not going to focus on addressing concerns too much on that end.
But the real crux of the matter is that the tech giant is working hard on rolling out improvements for AI Overviews that explain how to ensure improvements are done in the future.

For instance, it would like to design enhanced detection systems for queries that don’t make sense and therefore shouldn’t even be allowed to roll out an AI Overview. Secondly, it wishes to update current mechanisms in place so that it can limit user-produced content in replies providing misleading information.

And lastly, it hopes to instill more safeguards for sensitive topics linked to the world of health and current affairs. This means limiting AI Overviews for such topics that could have serious consequences.
The blog post ended with a commitment from Google’s side to continue improving and strengthening safeguards, hoping they can get the right feedback. So for now, AI Overviews are not going anywhere. But what we can hope is that they keep getting better as Google plans a massive launch to other regions of the world.

Image: DIW-Aigen

Read next:

New Analysis: Google AI Overviews Impacting 16% of Ecommerce Queries

• 42% of queries in the health vertical showed AI Overviews (AIOs).
Previous Post Next Post