Misinformation - Disinformation what you need to know in 2024

We rely on the internet to find information on almost everything, but how much of this information is accurate?

Misinformation is inaccurate information posted in error.

Disinformation is inaccurate information deliberately posted to mislead so as to influence thoughts and actions.

Malinformation is information that is true and posted to influence opinions and or actions,  usually by taking it out of context.    The posting of a photo with a misleading caption.  The misuse of private info – Phishing.  There are many examples in political communication.

Satire and parody can be classified as misinformation – but there is no intent to harm – it is assumed the audience understands the information is distorted for entertainment.

Then there are biased communications – for example media articles and books written from a particular perspective, with no mention of other viewpoints,  are misleading. They are essentially misinformation.   Likewise material that provides no context can also be misleading.

Unverified reports – Many reports appear in the main and social media with no verification – no reference to the source of the information.  This is true of some product claims, as well as some research and stories that are in fact rumor. In  the extreme there are conspiracy theories.  Depending on the intent, these reports can be misinformation or disinformation. Many who pass on such information are unaware of its potential falseness.

Propaganda is a deliberate attempt to strongly push a particular point of view – either with no reference to opposing views, or by claiming they are completely false.  This is disinformation

Fake news is by definition deliberate falsification – disinformation. It is intended to influence and manipulate the opinion of its audience.

It includes clickbait – sensational headlines to attract attention.  Misuse of maps, charts, data tables, photos and other graphics, false attribution of authorship or source of information are also used.  More recently, outright fabrications of images and information have appeared and entire fake websites.

The rolke of AI?  Generative AI infortunately has a tendency to produce false information by itself – termed AI hallucination.

Disinformation has always been around but the internet has made its creation and distribution exponentially faster and easier.

The majority – over 60% of the world population is online.  The total amount of content has exploded.   A huge percentage of this content is shared selectively, with bias – fuelled by algorithms designed to maximise user engagement for financial gain (advertising income), or for political purposes.  As a result, the reality that the internet audience perceives is invariably skewed.

Misinformation and disinformation have significant effects on society and individuals.  At an individual level it can cause loss of reputation,  adverse effects on health and financial loss from hoaxes and scams.  At the society level it can result in loss of confidence in public institutions and service, as well as impacting political agendas and elections.

GDI The Global Disinformation Index publishes research on disinformation by country, disinformation risk of news publishers and websites, and disinformation published on particular events and topics.

A 2023 Statista survey of  adults in the US found that up to 66% reporting seeing false information on line daily. The older the respondent the more likely they were to report this.

There are strong political and commercial incentives to distribute misleading content that promotes particular interests, and denigrates the competition.

The advert-funded model of much of the web facilitates and has boosted the creation of misleading information.    In particular the automated advertising auctions reward sites with many views,  advert impressions and clicks.  This is an incentive to produce, often misleading, sensational content that will attract users – “clickbait” .

Much of the enormous proliferation of content and disinformation on the web is due to Artificial Intelligence (AI).  AI has provided ways to easily generate text and manipulate it, as well as to manipulate images and audio visual media. Using AI actors can create realistic media that are similar, but not identical to, original content.

On the distribution side AI is deployed by tech platforms to grow user engagement. This is because of their need to reach as many users as possible to support the advertising revenue they depend on.

These algorithms have contributed to the extensive sharing and dissemination of misinformation. The algorithms promote the more contentious and sensational content to increase user engagement. Platforms track user behaviour across the internet using ‘Cookies’ embedded in sites users visit, as well as the particular fingerprint of a user’s browser. This browsing history is used to deliver search results in line with the user’s previous searches and interests.

Because sensational content is more popular,  a feedback loop is created – promoting the creation and consumption of even more such content.  In this way the algorithms affect the results of search, as well as material that is presented passively on websites.

The deployment of bots (software robots) by those with agendas also contributes to the dissemination of  disinformation.  Bots operate on social media platforms posting and sharing content posing as human contributors. They use fully or semi -automated fake user accounts.

The key questions

1. is AI effective in detecting  mis/disinformation?
2. is it appropriate to use AI  –  since AI is itself responsible for much of the proliferation of misinformation and disinformation?

Automated detection of false information requires language analysis.

The huge amount of information to be reviewed is itself a challenge –in large part due to the use of bots to deliberately create and distribute false information.  Much of this information is unstructured – especially that in social media.  This unstructured data has to be extracted into meaningful format using text mining methods.  Sentences are split up and key words identified,  ready for analysis.

Natural Language Processing (NLP) is automated analysis based on statistical frequency of word occurrences and relationships – proximity to other words.

Text is complex – the same message can be expressed in many different ways.

AI can identify simple sentences. When more complex text is involved it is has difficulty.  It is unable to handle nuances due to the use of sarcasm and cultural variations in language.

Therefore classification by textual analysis programs using natural language processing algorithms is not very effective.

Machine learning programmes can learn from humans who can flag a piece of text as true or false.  Then the program can review the characteristics of the text, find patterns and make classification decisions based on correlations.  This semi supervised machine learning method can enhance accuracy.

In the context of disinformation, fact-checking refers to the process of verifying information to confirm the accuracy of purported fact-based statements.

When humans fact check information they look at the context as well as the words.  There are many aspects of context – time frame, historical background, the individuals involved, locations, and many other factors that are related to events such as weather at the time.

Humans analyse the evidence,  assess the reliability and validity of the source of the information, and cross check the claims with multiple other sources.

AI needs to be trained as to what is true versus false. This means training material must be provided, and humans have to provide guidance on this,  at least initially.

Politifact and others have compiled databases of fact checks.  Currently these are very limited in terms of the numbers of checks. whereas AI needs very large numbers of records (millions) to produce meaningful results.

A further challenge is that for effective training of algorithms such databases need to have balanced examples of both true and false information on a wide range of topics.

Deep fake audio- visual material can only be detected by comparison with the original material and  a complete record of a person’s activity.

Many news media writers now provide opinion pieces from biased partisan perspectives, and the media itself tends to align with a preferred political agenda.  This bias is not transparent.

Ideally journalism organizes information for the public in a way that promotes the public interest. It should be free from bias with sources of information clearly provided, and points of contention identified. The media should be  a forum for the public to hold power to account. It should present news in a balanced way, without overblowing or under emphasizing threats and risks.

Stories should be based purely on the facts, supported by evidence. The language should be neutral – without emotive or positive/negative adjectives.  Where there are two or more sides to a story, the report should cover all perspectives equally.

Some of the major tech platforms have a code of practice for detecting and managing false information. However,  there is no access to their data and algorithms,  and therefore no way of auditing it.  Currently there seem to be no real protections for basic rights of individuals.

Reducing the reliance of platforms on advertising would reduce the amount of sensational and false content designed to attract user attention, and thus advertising revenue.  However introducing pay for service models would be very difficult, with ethical issues of its own.

Three key strategies fall under this heading.  Debunking and Innoculation (prebunking) and Nudging

Debunking is the correction of misinformation. It is most effective when the explanation includes – why it may be deemed correct, the incorrect information is highlighted,  and the reasons why it is wrong are provided,  together with the  facts – the alternative explanation, and why it is correct.

However it may not convince those who hold strong views.  Fact checking is labour intensive and the debunking of messages may not stay top of mind,  so effects diminish over time.

Pre-bunking  uses the psychological theory of Innoculation. The technique is to expose individuals to a false claim and counter arguments before they encounter the related disinformation.  ” You may see reports that …. but be aware that this is not true because …”.    The suggestion is that warning users of the possibility of misinformation will improve their awareness and judgment of misinformation generated by chatbots and other malign actors.

In accordance with innoculation theory – booster messages at intervals help maintain the effect.  This strategy is thought more likely to be effective than debunking, and easier to apply at scale.

Nudging

Nudges are prompts to consider the authenticity of information before sharing it.  They can be specific,  calling attention to accuracy and truth, as well as calling attention to community standards for sharing information.  The limited research has shown mixed results to date.

The Unesco Media Literacy Initiative has listed the key knowledge and skills people need in order to  understand  their  information  needs,  and  to locate, retrieve and evaluate the quality of the content they can access in a digital world.

It is recommended that each piece of content should contain information on

• how it was produced
• sponsors if any
• where it is available
• the target audience

This would enable those who access the information to better assess its validity.  Currently only some academic research studies provide this information.

For individuals and organizations Critical Thinking skills are now one of the most important competencies.

Critical thinking skills are used in Problem Solving and consist of;

• Interpretation,
• Analysis
• Reasoning
• Evaluation